feat(deploy): production docker compose stack + Gitea CI image build

Adds a complete deployment surface for production:

  Dockerfile               multi-stage 3.12-slim build, collectstatic
                           baked into the image, runs as non-root mnemosyne
                           uid/gid 1000.
  docker/entrypoint.sh     dispatches `web | mcp | worker | beat | migrate
                           | setup | shell` from a single image, so every
                           service in compose runs the same artifact.
  docker-compose.yaml      five services: static-init (one-shot copies
                           statics into the shared volume on every up),
                           web (gunicorn), mcp (uvicorn), worker (celery),
                           nginx. External services (Postgres, Neo4j,
                           RabbitMQ, S3, Memcached, embedder, reranker)
                           reached over the 10.10.0.0/24 internal network
                           and configured via mnemosyne/.env.
  nginx/mnemosyne.conf     reverse proxy: /library/* and /admin/* → web,
                           /mcp/* → mcp, /static/* → volume, /metrics
                           internal-network-only (127/8 + RFC1918), /healthz
                           proxies to /mcp/health for liveness probes.
  .gitea/workflows/        CVE scan + image build, image pushed to
                           git.helu.ca/r/mnemosyne. Trivy scans pyproject
                           extras (dev/test/lint/docs) and the built image.
  pyproject.toml           adds [test], [lint], [docs] extras so the CI
                           pip-compile step has something to resolve.

README documents the bring-up flow (`docker compose run --rm web migrate`,
then `setup`, then `up -d`), day-to-day commands, and the env-var values
that need adjusting for production (DEBUG=False, KVDB_LOCATION pointing
at the external memcached, AWS keys filled in, etc.).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

.gitea/workflows/cve-scan-docker-build.yml

@@ -0,0 +1,120 @@
+name: CVE Scan & Docker Build
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+env:
+  REGISTRY: git.helu.ca
+  IMAGE_NAME: ${{ gitea.repository }}
+  TRIVY_SEVERITY: MEDIUM,HIGH,CRITICAL
+  TRIVY_NO_PROGRESS: "true"
+  TRIVY_DISABLE_VEX_NOTICE: "true"
+
+jobs:
+  security-scan:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Install Trivy
+        run: |
+          curl -sfL https://raw.githubusercontent.com/aquasecurity/trivy/main/contrib/install.sh | sudo sh -s -- -b /usr/local/bin
+          trivy --version
+
+      - name: Resolve full dependency set (incl. dev/test/lint/docs extras)
+        run: |
+          python3 -m venv /tmp/scanenv
+          /tmp/scanenv/bin/pip install --quiet pip-tools
+          /tmp/scanenv/bin/pip-compile pyproject.toml \
+            --extra dev --extra test --extra lint --extra docs \
+            -o requirements.txt --no-header --quiet --allow-unsafe
+          echo "Resolved $(grep -cv '^\s*\(#\|$\)' requirements.txt) pinned packages."
+
+      - name: Scan Python dependencies for CVEs
+        run: |
+          trivy fs \
+            --scanners vuln \
+            --severity ${TRIVY_SEVERITY} \
+            --format table \
+            --exit-code 0 \
+            requirements.txt
+
+      - name: Scan repository for secrets
+        run: |
+          trivy fs \
+            --scanners secret \
+            --format table \
+            --exit-code 0 \
+            .
+
+  build-and-push:
+    runs-on: ubuntu-latest
+    needs: security-scan
+    if: always()
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to Gitea Container Registry
+        uses: docker/login-action@v3
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ gitea.actor }}
+          password: ${{ secrets.PACKAGE_TOKEN }}
+
+      - name: Extract metadata for Docker
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
+          tags: |
+            type=ref,event=branch
+            type=ref,event=pr
+            type=sha,prefix=
+            type=raw,value=latest,enable=${{ gitea.ref == 'refs/heads/main' }}
+
+      - name: Build and push Docker image
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          file: Dockerfile
+          push: true
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max
+
+      - name: Install Trivy
+        run: |
+          curl -sfL https://raw.githubusercontent.com/aquasecurity/trivy/main/contrib/install.sh | sudo sh -s -- -b /usr/local/bin
+          trivy --version
+
+      - name: Scan built Docker image (OS + Python + system libs)
+        run: |
+          IMAGE_TAG=$(echo "${{ steps.meta.outputs.tags }}" | head -n1)
+          echo "🔍 Scanning image: ${IMAGE_TAG}"
+          trivy image \
+            --scanners vuln \
+            --severity ${TRIVY_SEVERITY} \
+            --format table \
+            --pkg-types os,library \
+            --exit-code 0 \
+            "${IMAGE_TAG}"
+
+      - name: Scan built Docker image for misconfigurations
+        continue-on-error: true
+        run: |
+          IMAGE_TAG=$(echo "${{ steps.meta.outputs.tags }}" | head -n1)
+          trivy image \
+            --scanners misconfig \
+            --severity ${TRIVY_SEVERITY} \
+            --format table \
+            --exit-code 0 \
+            "${IMAGE_TAG}"

Dockerfile

@@ -0,0 +1,93 @@
+# =============================================================================
+# Mnemosyne — production image
+# =============================================================================
+# Multi-stage:
+#   builder  installs Python deps and runs `collectstatic` once.
+#   runtime  copies only the artifacts the running process needs.
+#
+# The same image runs three different processes (Django web, MCP server,
+# Celery worker) — the compose file picks the command per service.
+# =============================================================================
+
+# ── Stage 1: builder ────────────────────────────────────────────────────────
+FROM python:3.12-slim AS builder
+
+# Build deps for psycopg, PyMuPDF, Pillow, cryptography, etc.
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        build-essential \
+        libpq-dev \
+        libffi-dev \
+        libssl-dev \
+        libjpeg-dev \
+        zlib1g-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    PIP_NO_CACHE_DIR=1 \
+    PIP_DISABLE_PIP_VERSION_CHECK=1
+
+WORKDIR /build
+
+# Install dependencies first (better layer caching).
+COPY pyproject.toml README.md ./
+COPY mnemosyne/ ./mnemosyne/
+
+RUN pip install --upgrade pip \
+    && pip install .
+
+# Bake static files into the image. The env vars below are build-time-only
+# stubs needed for settings.py to import without real infrastructure — they
+# never reach the runtime image because this is the builder stage.
+# Inlined into the RUN command (rather than ENV/ARG) so static analysis
+# tools (Trivy) don't flag them as baked-in secrets.
+ENV DJANGO_SETTINGS_MODULE=mnemosyne.settings \
+    DEBUG=False \
+    USE_LOCAL_STORAGE=True \
+    APP_DB_NAME=collectstatic \
+    APP_DB_USER=collectstatic
+
+WORKDIR /build/mnemosyne
+RUN SECRET_KEY=collectstatic-stub \
+    APP_DB_PASSWORD=collectstatic-stub \
+    python manage.py collectstatic --noinput --clear
+
+# ── Stage 2: runtime ────────────────────────────────────────────────────────
+FROM python:3.12-slim AS runtime
+
+# Runtime libs for psycopg + PyMuPDF + Pillow + cryptography.
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        libpq5 \
+        libjpeg62-turbo \
+        zlib1g \
+        libssl3 \
+        ca-certificates \
+    && rm -rf /var/lib/apt/lists/*
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    DJANGO_SETTINGS_MODULE=mnemosyne.settings \
+    PATH=/usr/local/bin:$PATH
+
+# Copy installed packages from the builder.
+COPY --from=builder /usr/local/lib/python3.12/site-packages /usr/local/lib/python3.12/site-packages
+COPY --from=builder /usr/local/bin /usr/local/bin
+
+# Application code + collected statics.
+WORKDIR /app
+COPY --from=builder /build/mnemosyne /app
+COPY docker/entrypoint.sh /usr/local/bin/entrypoint.sh
+RUN chmod +x /usr/local/bin/entrypoint.sh
+
+# Non-root user for everything that runs in this image. uid:gid 1000:1000
+# matches the convention for a single-application container.
+RUN groupadd --gid 1000 mnemosyne \
+    && useradd --uid 1000 --gid mnemosyne --home /app --no-create-home --shell /sbin/nologin mnemosyne \
+    && mkdir -p /app/media /app/logs \
+    && chown -R mnemosyne:mnemosyne /app
+USER mnemosyne
+
+# The compose file overrides this per service. Default = Django web.
+EXPOSE 8000 22091
+ENTRYPOINT ["entrypoint.sh"]
+CMD ["web"]

README.md

@@ -37,11 +37,14 @@ This **content-type awareness** flows through every layer: chunking strategy, em
 | Library | Example Content | Multimodal? | Graph Relationships |
 |---------|----------------|-------------|-------------------|
 | **Fiction** | Novels, short stories | Cover art | Author → Book → Character → Theme |
+| **Nonfiction** | History, biography, science writing | Photos, charts | Author → Work → Topic → Person/Place |
 | **Technical** | Textbooks, manuals, docs | Diagrams, screenshots | Product → Manual → Section → Procedure |
 | **Music** | Lyrics, liner notes | Album artwork | Artist → Album → Track → Genre |
 | **Film** | Scripts, synopses | Stills, posters | Director → Film → Scene → Actor |
 | **Art** | Descriptions, catalogs | The artwork itself | Artist → Piece → Style → Movement |
-| **Journals** | Personal entries | Photos | Date → Entry → Topic → Person/Place |
+| **Journal** | Personal entries, plans, observations | Photos | Date → Entry → Topic → Person/Place |
+| **Business** | Proposals, marketing, strategy | Logos, charts | Client → Engagement → Deliverable |
+| **Finance** | Statements, tax, market commentary | Charts, statement scans | Account → Instrument → Period |
 
 ## Search Pipeline
 
@@ -55,32 +58,160 @@ Query → Vector Search (Neo4j) + Graph Traversal (Cypher) + Full-Text Search
 
 Mnemosyne's RAG pipeline architecture is inspired by [Spelunker](https://git.helu.ca/r/spelunker), an enterprise RFP response platform. The proven patterns — hybrid search, two-stage RAG (responder + reviewer), citation-based retrieval, and async document processing — are carried forward and enhanced with multimodal capabilities and knowledge graph relationships.
 
-## Running Celery Workers
+## Running Mnemosyne
 
-Mnemosyne uses Celery with RabbitMQ for async document embedding. From the `mnemosyne/` directory:
+Mnemosyne runs as three cooperating processes: the Django web app (REST API + admin), the MCP server (LLM-facing tools), and one or more Celery workers (async embedding + ingest). All three read configuration from `mnemosyne/.env` (copy from `mnemosyne/.env example` and fill in secrets).
+
+Hosts in the Ouranos lab:
+- **Postgres** — `portia.incus:5432` (Django ORM: users, IngestJob)
+- **Neo4j** — `ariel.incus:25554` (knowledge graph + vectors)
+- **RabbitMQ** — `oberon.incus:5672` (Celery broker)
+- **MinIO** — `nyx.helu.ca:8555` (S3-compatible; `mnemosyne-content` and `daedalus` buckets)
+- **Memcached** — `127.0.0.1:11211` (task progress)
+
+### One-time setup
 
 ```bash
-# Development — single worker, all queues
+cd mnemosyne/
+python manage.py migrate                       # Apply Django ORM migrations
+python manage.py setup_neo4j_indexes           # Create Neo4j vector + full-text indexes
+python manage.py load_library_types            # Load LIBRARY_TYPE_DEFAULTS into Neo4j
+```
+
+### Start the web app
+
+The Django REST API serves `/library/api/*` (libraries, collections, items, search, workspaces, ingest) and Django admin. Use Gunicorn in production; `runserver` for dev.
+
+```bash
+cd mnemosyne/
+
+# Development
+python manage.py runserver 0.0.0.0:8000
+
+# Production
+gunicorn --bind 0.0.0.0:8000 --workers 3 mnemosyne.wsgi:application
+```
+
+### Start the MCP server
+
+The MCP server exposes the LLM-facing tools (`search`, `get_chunk`, `list_libraries`, `list_collections`, `list_items`, `get_health`) over Streamable HTTP at `/mcp` and SSE at `/mcp/sse`. Run as a separate Uvicorn process, on its own port, so it can be reverse-proxied or scaled independently of the Django app.
+
+```bash
+cd mnemosyne/
+
+# Single command: ASGI server hosting the FastMCP app
+uvicorn mnemosyne.asgi:app --host 0.0.0.0 --port 22091 --workers 1
+```
+
+The `mcp_server/asgi.py` mounts FastMCP at `/mcp` (Streamable HTTP) and `/mcp/sse` (SSE), with a `/mcp/health` JSON probe for HAProxy/Pallas.
+
+### Start a Celery worker
+
+A single worker that handles all queues (development) plus the focused command Daedalus depends on (the `embedding` queue, where the Daedalus ingest task lives).
+
+```bash
+cd mnemosyne/
+
+# Development — one worker, all queues
 celery -A mnemosyne worker -l info -Q celery,embedding,batch
 
-# Or skip workers entirely with eager mode (.env):
-CELERY_TASK_ALWAYS_EAGER=True
+# Production — embedding queue (handles Daedalus ingest + embed_item)
+celery -A mnemosyne worker -l info -Q embedding -c 1 -n embedding@%h
+
+# Production — batch queue (collection/library bulk operations)
+celery -A mnemosyne worker -l info -Q batch -c 2 -n batch@%h
+
+# Production — default queue (LLM validation, misc)
+celery -A mnemosyne worker -l info -Q celery -c 2 -n default@%h
 ```
 
-**Production — separate workers:**
-```bash
-celery -A mnemosyne worker -l info -Q embedding -c 1 -n embedding@%h    # GPU-bound embedding
-celery -A mnemosyne worker -l info -Q batch -c 2 -n batch@%h            # Batch orchestration
-celery -A mnemosyne worker -l info -Q celery -c 2 -n default@%h         # LLM API validation
-```
+Daedalus's `POST /library/api/ingest/` dispatches `library.tasks.ingest_from_daedalus` to the **embedding** queue. If you only run one worker, make sure it consumes `embedding` or that task will sit in the broker.
 
-**Scheduler & Monitoring:**
+To bypass workers in dev/test, set `CELERY_TASK_ALWAYS_EAGER=True` in `.env`.
+
+**Scheduler & monitoring (optional):**
 ```bash
 celery -A mnemosyne beat -l info            # Periodic task scheduler
 celery -A mnemosyne flower --port=5555      # Web monitoring UI
 ```
 
-See [Phase 2: Celery Workers & Scheduler](docs/PHASE_2_EMBEDDING_PIPELINE.md#celery-workers--scheduler) for full details on queues, reliability settings, and task progress tracking.
+See [Phase 2: Celery Workers & Scheduler](docs/PHASE_2_EMBEDDING_PIPELINE.md#celery-workers--scheduler) for queue tuning, reliability settings, and task progress tracking.
+
+### Daedalus integration endpoints
+
+These endpoints are used by the Daedalus FastAPI backend (HTTP Basic auth). All under `/library/api/`:
+
+| Method | Route | Purpose |
+|--------|-------|---------|
+| POST | `/workspaces/` | Create a workspace (idempotent on `workspace_id`); body: `{workspace_id, name, library_type, description?}` |
+| GET | `/workspaces/{workspace_id}/` | Workspace status (item/chunk counts) |
+| DELETE | `/workspaces/{workspace_id}/` | Delete workspace + reachable content; preserves shared concepts |
+| POST | `/ingest/` | Queue a file for ingestion + embedding |
+| GET | `/jobs/{job_id}/` | Poll ingest job status |
+| POST | `/jobs/{job_id}/retry/` | Re-dispatch a failed job |
+| GET | `/jobs/?status=&library_uid=` | List recent jobs |
+
+See [docs/mnemosyne_integration.md](docs/mnemosyne_integration.md) for the full Daedalus contract.
+
+## Production Deployment
+
+Production runs as four containers from a single image (built and pushed by [`.gitea/workflows/cve-scan-docker-build.yml`](.gitea/workflows/cve-scan-docker-build.yml) on every push to `main`):
+
+| Service | Role | Port |
+|---------|------|------|
+| `web` | Django REST API + admin (gunicorn) | internal :8000 |
+| `mcp` | FastMCP server (uvicorn) | internal :22091 |
+| `worker` | Celery worker — embedding/ingest/batch | — |
+| `nginx` | Reverse proxy + static files | host :23090 |
+
+Plus a one-shot `static-init` service that copies `/app/staticfiles` (baked into the image at build time via `collectstatic`) into the shared volume nginx reads from. It runs to completion on every `up`, so static-file changes propagate on each deploy without manual intervention.
+
+External services (NOT spun up by compose): Postgres on Portia, Neo4j on Ariel, RabbitMQ on Oberon, S3/MinIO on Nyx, Memcached, embedder + reranker. All reached over the internal 10.10.0.0/24 network. URLs and credentials live in `mnemosyne/.env`.
+
+### First-time bring-up
+
+```bash
+# Pull the image (or build locally with `docker compose build`)
+docker compose pull
+
+# DB migrations (one-shot)
+docker compose run --rm web migrate
+
+# Neo4j indexes + library_type defaults (one-shot)
+docker compose run --rm web setup
+
+# Bring the stack up
+docker compose up -d
+```
+
+### Day-to-day
+
+```bash
+docker compose ps                  # service status + health
+docker compose logs -f web         # tail web logs
+docker compose logs -f worker      # tail Celery worker logs
+docker compose restart mcp         # restart just the MCP server
+
+# After a new image is published:
+docker compose pull && docker compose up -d
+```
+
+### Things to verify in `mnemosyne/.env` before bringing up
+
+The development `.env` has a few values that need adjusting for production:
+
+- `DEBUG=False`
+- `USE_LOCAL_STORAGE=False` (already set; just confirm)
+- `KVDB_LOCATION=<external-memcached-host>:11211` — `127.0.0.1` does not resolve from inside containers
+- `AWS_ACCESS_KEY_ID` / `AWS_SECRET_ACCESS_KEY` filled in
+- `DAEDALUS_S3_*` filled in for cross-bucket reads from the Daedalus bucket
+- `ALLOWED_HOSTS` includes the public hostname HAProxy routes to (e.g. `mnemosyne.ouranos.helu.ca`)
+- `LLM_API_SECRETS_ENCRYPTION_KEY` set to a real Fernet key
+
+### Health probes
+
+- `GET http://nginx-host:23090/healthz` → proxies to `/mcp/health`, returns `{"status":"ok"}` when the MCP server is up
+- `GET http://nginx-host:23090/metrics` → Prometheus scrape endpoint, internal-network-only
 
 ## Architecture Note: Retrieval, Not Synthesis
 

docker-compose.yaml

@@ -0,0 +1,111 @@
+# =============================================================================
+# Mnemosyne — production deployment
+# =============================================================================
+# Four services, all from the same image:
+#   web      — Django REST API + admin (gunicorn, port 8000)
+#   mcp      — FastMCP server (uvicorn, port 22091)
+#   worker   — Celery worker (embedding/ingest/batch queues)
+#   nginx    — reverse proxy, public port 23090
+#
+# External services (NOT spun up here): Postgres on Portia, Neo4j on Ariel,
+# RabbitMQ on Oberon, S3/MinIO on Nyx, Memcached on its own host, embedder
+# and reranker on Nyx, smtp4dev on Oberon. All reached over the internal
+# 10.10.0.0/24 network.
+#
+# Run:
+#   docker compose up -d
+#   docker compose run --rm web migrate    # one-shot DB migrate
+#   docker compose run --rm web setup      # Neo4j indexes + library types
+# =============================================================================
+
+services:
+  # ── Static-file seeder: copies /app/staticfiles into the shared volume on
+  # every `up`. Runs once and exits. Without this, the named volume is only
+  # seeded the first time it's empty, so static updates between deploys
+  # would not propagate to nginx.
+  static-init:
+    image: git.helu.ca/r/mnemosyne:latest
+    command: ["sh", "-c", "cp -a /app/staticfiles/. /shared-static/"]
+    user: "0:0"
+    volumes:
+      - mnemosyne-static:/shared-static
+    restart: "no"
+
+  # ── Web app: Django REST API + admin ───────────────────────────────────────
+  web:
+    image: git.helu.ca/r/mnemosyne:latest
+    command: ["web"]
+    env_file: mnemosyne/.env
+    restart: unless-stopped
+    depends_on:
+      static-init:
+        condition: service_completed_successfully
+    volumes:
+      - mnemosyne-media:/app/media
+    expose:
+      - "8000"
+    healthcheck:
+      test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8000/admin/login/').read()"]
+      interval: 30s
+      timeout: 5s
+      retries: 3
+      start_period: 30s
+
+  # ── MCP server: FastMCP Streamable HTTP at /mcp/ ───────────────────────────
+  mcp:
+    image: git.helu.ca/r/mnemosyne:latest
+    command: ["mcp"]
+    env_file: mnemosyne/.env
+    restart: unless-stopped
+    volumes:
+      - mnemosyne-media:/app/media
+    expose:
+      - "22091"
+    healthcheck:
+      test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:22091/mcp/health').read()"]
+      interval: 30s
+      timeout: 5s
+      retries: 3
+      start_period: 30s
+
+  # ── Celery worker: embedding + ingest + batch queues ───────────────────────
+  worker:
+    image: git.helu.ca/r/mnemosyne:latest
+    command: ["worker"]
+    env_file: mnemosyne/.env
+    restart: unless-stopped
+    volumes:
+      - mnemosyne-media:/app/media
+    healthcheck:
+      test: ["CMD", "celery", "-A", "mnemosyne", "inspect", "ping", "-d", "celery@$$HOSTNAME"]
+      interval: 60s
+      timeout: 10s
+      retries: 3
+      start_period: 60s
+
+  # ── nginx: reverse proxy, public port 23090 ────────────────────────────────
+  nginx:
+    image: nginx:alpine
+    restart: unless-stopped
+    depends_on:
+      - web
+      - mcp
+    ports:
+      - "23090:80"
+    volumes:
+      - ./nginx/mnemosyne.conf:/etc/nginx/conf.d/default.conf:ro
+      - mnemosyne-static:/var/www/static:ro
+    healthcheck:
+      test: ["CMD", "wget", "-qO-", "http://localhost/healthz"]
+      interval: 30s
+      timeout: 5s
+      retries: 3
+
+volumes:
+  # Static files baked into the image at /app/staticfiles. The web service
+  # mounts this volume, populating it on first start; nginx reads from it.
+  mnemosyne-static:
+  # Local FileSystemStorage fallback. Production uses USE_LOCAL_STORAGE=False
+  # so this is mostly empty — kept for parity with dev and for any path
+  # that writes to MEDIA_ROOT directly.
+  mnemosyne-media:

docker/entrypoint.sh

@@ -0,0 +1,66 @@
+#!/bin/sh
+# Mnemosyne container entrypoint.
+#
+# The same image runs all three processes — the compose service supplies
+# `web`, `mcp`, `worker`, or `migrate` as CMD.
+
+set -e
+
+case "$1" in
+  web)
+    # Django REST API + admin (gunicorn → wsgi).
+    exec gunicorn \
+      --bind 0.0.0.0:8000 \
+      --workers "${GUNICORN_WORKERS:-3}" \
+      --access-logfile - \
+      --error-logfile - \
+      mnemosyne.wsgi:application
+    ;;
+
+  mcp)
+    # FastMCP over Streamable HTTP at /mcp/, mounted by mnemosyne.asgi.
+    exec uvicorn \
+      --host 0.0.0.0 \
+      --port 22091 \
+      --workers "${UVICORN_WORKERS:-1}" \
+      mnemosyne.asgi:app
+    ;;
+
+  worker)
+    # Celery worker covering embedding + ingest + batch + default queues.
+    # In production you may want to split these onto separate worker
+    # services for queue-level isolation; one process is fine to start.
+    exec celery -A mnemosyne worker \
+      --loglevel="${CELERY_LOG_LEVEL:-info}" \
+      --queues="${CELERY_QUEUES:-celery,embedding,batch}" \
+      --concurrency="${CELERY_CONCURRENCY:-2}"
+    ;;
+
+  beat)
+    # Celery scheduled tasks (only needed if/when periodic jobs are wired).
+    exec celery -A mnemosyne beat \
+      --loglevel="${CELERY_LOG_LEVEL:-info}"
+    ;;
+
+  migrate)
+    # One-shot DB migration runner — invoke before bringing services up
+    # for the first time or after a deploy.
+    exec python manage.py migrate --noinput
+    ;;
+
+  setup)
+    # One-shot init — Neo4j indexes + library_type seed data.
+    python manage.py setup_neo4j_indexes
+    python manage.py load_library_types
+    ;;
+
+  shell)
+    # Drop into the management shell for ad-hoc work.
+    exec python manage.py shell
+    ;;
+
+  *)
+    # Fall through: run whatever was passed (e.g. `manage.py <cmd>`).
+    exec "$@"
+    ;;
+esac

mnemosyne/.env example

@@ -40,6 +40,23 @@ AWS_S3_REGION_NAME=us-east-1
 # Set to True to use local FileSystemStorage instead of S3 (dev/test)
 USE_LOCAL_STORAGE=True
 
+# --- Daedalus S3 (cross-bucket reads for ingest) ---
+# Mnemosyne ingests files from the Daedalus S3 bucket. These vars
+# configure read access; the file is copied into AWS_STORAGE_BUCKET_NAME
+# (Mnemosyne's own bucket) by the ingest Celery task.
+DAEDALUS_S3_ENDPOINT_URL=
+DAEDALUS_S3_ACCESS_KEY_ID=
+DAEDALUS_S3_SECRET_ACCESS_KEY=
+DAEDALUS_S3_BUCKET_NAME=daedalus
+DAEDALUS_S3_REGION_NAME=us-east-1
+DAEDALUS_S3_USE_SSL=False
+DAEDALUS_S3_VERIFY=True
+
+# --- MCP Server ---
+# Set to False for internal-only deployments where the MCP transport is
+# already on a trusted network (10.10.0.0/24).
+MCP_REQUIRE_AUTH=True
+
 # --- Email (smtp4dev on Oberon) ---
 EMAIL_HOST=oberon.incus
 EMAIL_PORT=22025

mnemosyne/library/api/serializers.py

@@ -7,12 +7,23 @@ Serialize Neo4j neomodel nodes into JSON for the REST API.
 from rest_framework import serializers
 
 
+LIBRARY_TYPE_CHOICES = [
+    "fiction",
+    "nonfiction",
+    "technical",
+    "music",
+    "film",
+    "art",
+    "journal",
+    "business",
+    "finance",
+]
+
+
 class LibrarySerializer(serializers.Serializer):
     uid = serializers.CharField(read_only=True)
     name = serializers.CharField(max_length=200)
-    library_type = serializers.ChoiceField(
-        choices=["fiction", "nonfiction", "technical", "music", "film", "art", "journal"]
-    )
+    library_type = serializers.ChoiceField(choices=LIBRARY_TYPE_CHOICES)
     description = serializers.CharField(required=False, allow_blank=True, default="")
     chunking_config = serializers.JSONField(required=False, default=dict)
     embedding_instruction = serializers.CharField(
@@ -24,6 +35,7 @@ class LibrarySerializer(serializers.Serializer):
     llm_context_prompt = serializers.CharField(
         required=False, allow_blank=True, default=""
     )
+    workspace_id = serializers.CharField(read_only=True)
     created_at = serializers.DateTimeField(read_only=True)
 
 
@@ -90,12 +102,11 @@ class SearchRequestSerializer(serializers.Serializer):
     query = serializers.CharField(max_length=2000)
     library_uid = serializers.CharField(required=False, allow_blank=True)
     library_type = serializers.ChoiceField(
-        choices=[
-            "fiction", "nonfiction", "technical", "music", "film", "art", "journal",
-        ],
+        choices=LIBRARY_TYPE_CHOICES,
         required=False,
     )
     collection_uid = serializers.CharField(required=False, allow_blank=True)
+    workspace_id = serializers.CharField(required=False, allow_blank=True)
     search_types = serializers.ListField(
         child=serializers.ChoiceField(choices=["vector", "fulltext", "graph"]),
         required=False,
@@ -139,3 +150,73 @@ class SearchResponseSerializer(serializers.Serializer):
     reranker_used = serializers.BooleanField()
     reranker_model = serializers.CharField(allow_null=True)
     search_types_used = serializers.ListField(child=serializers.CharField())
+
+
+# --- Workspace lifecycle (Daedalus integration) ---
+
+
+class WorkspaceCreateSerializer(serializers.Serializer):
+    """Inbound payload for POST /api/v1/workspaces/."""
+
+    workspace_id = serializers.CharField(max_length=64)
+    name = serializers.CharField(max_length=200)
+    library_type = serializers.ChoiceField(choices=LIBRARY_TYPE_CHOICES)
+    description = serializers.CharField(required=False, allow_blank=True, default="")
+
+
+class WorkspaceStatusSerializer(serializers.Serializer):
+    """Outbound payload for workspace lifecycle endpoints."""
+
+    workspace_id = serializers.CharField()
+    library_uid = serializers.CharField()
+    name = serializers.CharField()
+    library_type = serializers.CharField()
+    description = serializers.CharField(allow_blank=True)
+    item_count = serializers.IntegerField()
+    chunk_count = serializers.IntegerField()
+    created_at = serializers.DateTimeField()
+
+
+# --- Ingest (Daedalus integration) ---
+
+
+class IngestRequestSerializer(serializers.Serializer):
+    """Inbound payload for POST /api/v1/library/ingest/."""
+
+    s3_key = serializers.CharField(max_length=500)
+    title = serializers.CharField(max_length=500)
+    library_uid = serializers.CharField(required=False, allow_blank=True)
+    workspace_id = serializers.CharField(required=False, allow_blank=True)
+    collection_uid = serializers.CharField(required=False, allow_blank=True)
+    file_type = serializers.CharField(required=False, allow_blank=True, default="")
+    file_size = serializers.IntegerField(required=False, default=0)
+    content_hash = serializers.CharField(max_length=64)
+    source = serializers.CharField(required=False, allow_blank=True, default="")
+    source_ref = serializers.CharField(required=False, allow_blank=True, default="")
+
+    def validate(self, data):
+        if not data.get("library_uid") and not data.get("workspace_id"):
+            raise serializers.ValidationError(
+                "Either library_uid or workspace_id is required."
+            )
+        return data
+
+
+class IngestJobSerializer(serializers.Serializer):
+    """Outbound payload for ingest job status."""
+
+    job_id = serializers.CharField(source="id")
+    item_uid = serializers.CharField(allow_blank=True)
+    library_uid = serializers.CharField()
+    status = serializers.CharField()
+    progress = serializers.CharField()
+    error = serializers.CharField(allow_null=True)
+    chunks_created = serializers.IntegerField()
+    concepts_extracted = serializers.IntegerField()
+    embedding_model = serializers.CharField(allow_blank=True)
+    content_hash = serializers.CharField(allow_blank=True)
+    source = serializers.CharField(allow_blank=True)
+    source_ref = serializers.CharField(allow_blank=True)
+    created_at = serializers.DateTimeField()
+    started_at = serializers.DateTimeField(allow_null=True)
+    completed_at = serializers.DateTimeField(allow_null=True)

mnemosyne/library/api/urls.py

@@ -4,7 +4,7 @@ URL patterns for the library DRF API.
 
 from django.urls import path
 
-from . import views
+from . import views, workspaces
 
 app_name = "library-api"
 
@@ -28,4 +28,16 @@ urlpatterns = [
     # Concepts (Phase 3)
     path("concepts/", views.concept_list, name="concept-list"),
     path("concepts/<str:uid>/graph/", views.concept_graph, name="concept-graph"),
+    # Workspaces (Daedalus integration)
+    path("workspaces/", workspaces.workspace_create, name="workspace-create"),
+    path(
+        "workspaces/<str:workspace_id>/",
+        workspaces.workspace_detail_or_delete,
+        name="workspace-detail",
+    ),
+    # Ingest (Daedalus integration)
+    path("ingest/", views.ingest_create, name="ingest-create"),
+    path("jobs/", views.ingest_job_list, name="ingest-job-list"),
+    path("jobs/<str:job_id>/", views.ingest_job_detail, name="ingest-job-detail"),
+    path("jobs/<str:job_id>/retry/", views.ingest_job_retry, name="ingest-job-retry"),
 ]

mnemosyne/library/api/views.py

@@ -21,6 +21,8 @@ from library.content_types import get_library_type_config
 from .serializers import (
     CollectionSerializer,
     ConceptSerializer,
+    IngestJobSerializer,
+    IngestRequestSerializer,
     ItemSerializer,
     LibrarySerializer,
     SearchRequestSerializer,
@@ -456,6 +458,7 @@ def search(request):
         library_uid=data.get("library_uid") or None,
         library_type=data.get("library_type") or None,
         collection_uid=data.get("collection_uid") or None,
+        workspace_id=data.get("workspace_id") or None,
         search_types=data.get("search_types", ["vector", "fulltext", "graph"]),
         limit=data.get("limit", getattr(django_settings, "SEARCH_DEFAULT_LIMIT", 20)),
         vector_top_k=getattr(django_settings, "SEARCH_VECTOR_TOP_K", 50),
@@ -487,6 +490,7 @@ def search_vector(request):
         library_uid=data.get("library_uid") or None,
         library_type=data.get("library_type") or None,
         collection_uid=data.get("collection_uid") or None,
+        workspace_id=data.get("workspace_id") or None,
         search_types=["vector"],
         limit=data.get("limit", 20),
         vector_top_k=getattr(django_settings, "SEARCH_VECTOR_TOP_K", 50),
@@ -517,6 +521,7 @@ def search_fulltext(request):
         library_uid=data.get("library_uid") or None,
         library_type=data.get("library_type") or None,
         collection_uid=data.get("collection_uid") or None,
+        workspace_id=data.get("workspace_id") or None,
         search_types=["fulltext"],
         limit=data.get("limit", 20),
         fulltext_top_k=getattr(django_settings, "SEARCH_FULLTEXT_TOP_K", 30),
@@ -635,3 +640,196 @@ def concept_graph(request, uid):
             {"detail": f"Failed: {exc}"},
             status=status.HTTP_500_INTERNAL_SERVER_ERROR,
         )
+
+
+# ---------------------------------------------------------------------------
+# Ingest API (Daedalus integration)
+# ---------------------------------------------------------------------------
+
+
+@api_view(["POST"])
+@permission_classes([IsAuthenticated])
+def ingest_create(request):
+    """
+    Accept a file (already in S3) for ingestion + embedding.
+
+    Daedalus calls this after committing a file to its own S3 bucket.
+    We resolve the target Library (by workspace_id or library_uid),
+    enforce idempotency on (library, source_ref, content_hash), create
+    an IngestJob row, and dispatch a Celery task.
+
+    Idempotency:
+    - Same source_ref + same content_hash → return existing completed job
+      (no new task dispatched).
+    - Same source_ref + different content_hash → dispatch a new task that
+      will supersede the prior Item.
+    - New source_ref → fresh ingest.
+    """
+    import uuid
+
+    from library.models import IngestJob, Library
+    from library.tasks import ingest_from_daedalus
+
+    serializer = IngestRequestSerializer(data=request.data)
+    serializer.is_valid(raise_exception=True)
+    data = serializer.validated_data
+
+    # --- Resolve target Library ---
+    workspace_id = data.get("workspace_id") or ""
+    library_uid = data.get("library_uid") or ""
+    if workspace_id:
+        try:
+            lib = Library.nodes.get(workspace_id=workspace_id)
+        except Library.DoesNotExist:
+            return Response(
+                {"detail": f"Workspace '{workspace_id}' not registered."},
+                status=status.HTTP_404_NOT_FOUND,
+            )
+    else:
+        try:
+            lib = Library.nodes.get(uid=library_uid)
+        except Library.DoesNotExist:
+            return Response(
+                {"detail": f"Library '{library_uid}' not found."},
+                status=status.HTTP_404_NOT_FOUND,
+            )
+
+    # --- Idempotency check on (library, source_ref, content_hash) ---
+    source_ref = data.get("source_ref") or ""
+    content_hash = data["content_hash"]
+
+    if source_ref:
+        existing = (
+            IngestJob.objects
+            .filter(
+                library_uid=lib.uid,
+                source_ref=source_ref,
+                content_hash=content_hash,
+                status__in=["pending", "processing", "completed"],
+            )
+            .order_by("-created_at")
+            .first()
+        )
+        if existing is not None:
+            logger.info(
+                "Ingest idempotent hit job_id=%s source_ref=%s status=%s",
+                existing.id, source_ref, existing.status,
+            )
+            return Response(
+                IngestJobSerializer(existing).data,
+                status=status.HTTP_200_OK,
+            )
+
+    # --- Create job + dispatch ---
+    job = IngestJob.objects.create(
+        id=f"job_{uuid.uuid4().hex[:24]}",
+        library_uid=lib.uid,
+        s3_key=data["s3_key"],
+        title=data["title"],
+        file_type=data.get("file_type", ""),
+        file_size=data.get("file_size", 0),
+        content_hash=content_hash,
+        source=data.get("source", ""),
+        source_ref=source_ref,
+        collection_uid=data.get("collection_uid", ""),
+        status="pending",
+        progress="queued",
+    )
+
+    try:
+        async_result = ingest_from_daedalus.delay(job.id)
+        job.celery_task_id = async_result.id
+        job.save(update_fields=["celery_task_id"])
+    except Exception as exc:
+        logger.error("Failed to dispatch ingest task job_id=%s: %s", job.id, exc)
+        job.status = "failed"
+        job.error = f"dispatch failed: {exc}"
+        job.save(update_fields=["status", "error"])
+        return Response(
+            IngestJobSerializer(job).data,
+            status=status.HTTP_500_INTERNAL_SERVER_ERROR,
+        )
+
+    logger.info(
+        "Ingest dispatched job_id=%s library_uid=%s source_ref=%s task_id=%s",
+        job.id, lib.uid, source_ref, job.celery_task_id,
+    )
+
+    return Response(
+        IngestJobSerializer(job).data,
+        status=status.HTTP_202_ACCEPTED,
+    )
+
+
+@api_view(["GET"])
+@permission_classes([IsAuthenticated])
+def ingest_job_detail(request, job_id):
+    """Get the current status of an IngestJob."""
+    from library.models import IngestJob
+
+    try:
+        job = IngestJob.objects.get(pk=job_id)
+    except IngestJob.DoesNotExist:
+        return Response(
+            {"detail": "Job not found."}, status=status.HTTP_404_NOT_FOUND
+        )
+
+    return Response(IngestJobSerializer(job).data)
+
+
+@api_view(["POST"])
+@permission_classes([IsAuthenticated])
+def ingest_job_retry(request, job_id):
+    """Re-dispatch a failed IngestJob."""
+    from library.models import IngestJob
+    from library.tasks import ingest_from_daedalus
+
+    try:
+        job = IngestJob.objects.get(pk=job_id)
+    except IngestJob.DoesNotExist:
+        return Response(
+            {"detail": "Job not found."}, status=status.HTTP_404_NOT_FOUND
+        )
+
+    if job.status not in ("failed", "completed"):
+        return Response(
+            {"detail": f"Job is currently {job.status}; cannot retry."},
+            status=status.HTTP_409_CONFLICT,
+        )
+
+    job.status = "pending"
+    job.progress = "queued"
+    job.error = ""
+    job.save(update_fields=["status", "progress", "error"])
+
+    async_result = ingest_from_daedalus.delay(job.id)
+    job.celery_task_id = async_result.id
+    job.save(update_fields=["celery_task_id"])
+
+    logger.info("Ingest retry dispatched job_id=%s task_id=%s", job.id, async_result.id)
+    return Response(IngestJobSerializer(job).data, status=status.HTTP_202_ACCEPTED)
+
+
+@api_view(["GET"])
+@permission_classes([IsAuthenticated])
+def ingest_job_list(request):
+    """List recent IngestJob rows, optionally filtered by status / library_uid."""
+    from library.models import IngestJob
+
+    qs = IngestJob.objects.all()
+    status_filter = request.query_params.get("status")
+    library_uid = request.query_params.get("library_uid")
+    limit = min(int(request.query_params.get("limit", 50)), 200)
+
+    if status_filter:
+        qs = qs.filter(status=status_filter)
+    if library_uid:
+        qs = qs.filter(library_uid=library_uid)
+
+    jobs = list(qs.order_by("-created_at")[:limit])
+    return Response(
+        {
+            "jobs": IngestJobSerializer(jobs, many=True).data,
+            "count": len(jobs),
+        }
+    )

mnemosyne/library/api/workspaces.py

@@ -0,0 +1,217 @@
+"""
+Workspace lifecycle endpoints for the Daedalus integration.
+
+A "workspace" in Mnemosyne is a Library scoped to a Daedalus workspace UUID.
+It uses the same Library node as a global library; the difference is that
+`workspace_id` is set, and search must filter on it.
+
+These endpoints are called by the Daedalus backend (HTTP Basic auth as
+the `daedalus-service` user). Daedalus owns the workspace_id; Mnemosyne
+just persists what Daedalus tells it.
+"""
+
+import logging
+
+from neomodel import db
+from rest_framework import status
+from rest_framework.decorators import api_view, permission_classes
+from rest_framework.permissions import IsAuthenticated
+from rest_framework.response import Response
+
+from library.content_types import get_library_type_config
+
+from .serializers import WorkspaceCreateSerializer, WorkspaceStatusSerializer
+
+logger = logging.getLogger(__name__)
+
+
+def _serialize_workspace(lib):
+    """Build a WorkspaceStatus payload from a Library node + chunk/item counts."""
+    counts, _ = db.cypher_query(
+        "MATCH (l:Library {workspace_id: $wsid}) "
+        "OPTIONAL MATCH (l)-[:CONTAINS]->(:Collection)-[:CONTAINS]->(i:Item) "
+        "OPTIONAL MATCH (i)-[:HAS_CHUNK]->(c:Chunk) "
+        "RETURN count(DISTINCT i) AS item_count, count(DISTINCT c) AS chunk_count",
+        {"wsid": lib.workspace_id},
+    )
+    item_count = counts[0][0] if counts else 0
+    chunk_count = counts[0][1] if counts else 0
+
+    return {
+        "workspace_id": lib.workspace_id,
+        "library_uid": lib.uid,
+        "name": lib.name,
+        "library_type": lib.library_type,
+        "description": lib.description or "",
+        "item_count": item_count,
+        "chunk_count": chunk_count,
+        "created_at": lib.created_at,
+    }
+
+
+@api_view(["POST"])
+@permission_classes([IsAuthenticated])
+def workspace_create(request):
+    """
+    Create a workspace Library, idempotently.
+
+    A POST with a `workspace_id` already in use returns the existing
+    workspace (200) — not an error. The library_type is frozen at first
+    create; subsequent calls are not allowed to change it.
+    """
+    from library.models import Library
+
+    serializer = WorkspaceCreateSerializer(data=request.data)
+    serializer.is_valid(raise_exception=True)
+    data = serializer.validated_data
+
+    # Idempotent path: workspace already exists.
+    try:
+        existing = Library.nodes.get(workspace_id=data["workspace_id"])
+    except Library.DoesNotExist:
+        existing = None
+
+    if existing is not None:
+        if existing.library_type != data["library_type"]:
+            return Response(
+                {
+                    "detail": (
+                        "library_type is immutable for an existing workspace "
+                        f"(have '{existing.library_type}', "
+                        f"got '{data['library_type']}')."
+                    )
+                },
+                status=status.HTTP_409_CONFLICT,
+            )
+        logger.info(
+            "Workspace already exists workspace_id=%s library_uid=%s",
+            data["workspace_id"], existing.uid,
+        )
+        return Response(
+            WorkspaceStatusSerializer(_serialize_workspace(existing)).data,
+            status=status.HTTP_200_OK,
+        )
+
+    defaults = get_library_type_config(data["library_type"])
+    lib = Library(
+        name=data["name"],
+        library_type=data["library_type"],
+        description=data.get("description", ""),
+        workspace_id=data["workspace_id"],
+        chunking_config=defaults["chunking_config"],
+        embedding_instruction=defaults["embedding_instruction"],
+        reranker_instruction=defaults["reranker_instruction"],
+        llm_context_prompt=defaults["llm_context_prompt"],
+    )
+    lib.save()
+    logger.info(
+        "Workspace created workspace_id=%s library_uid=%s library_type=%s",
+        data["workspace_id"], lib.uid, lib.library_type,
+    )
+
+    return Response(
+        WorkspaceStatusSerializer(_serialize_workspace(lib)).data,
+        status=status.HTTP_201_CREATED,
+    )
+
+
+@api_view(["GET", "DELETE"])
+@permission_classes([IsAuthenticated])
+def workspace_detail_or_delete(request, workspace_id):
+    """
+    GET: return workspace status (item/chunk counts, metadata).
+
+    DELETE: delete the workspace Library and everything reachable AND unique
+    to it. Concept-safe: orphan-only Concept GC happens at the end.
+    Concepts referenced by other libraries (workspace or global) are preserved.
+    """
+    from library.models import Library
+
+    if request.method == "GET":
+        try:
+            lib = Library.nodes.get(workspace_id=workspace_id)
+        except Library.DoesNotExist:
+            return Response(
+                {"detail": "Workspace not found."},
+                status=status.HTTP_404_NOT_FOUND,
+            )
+
+        return Response(WorkspaceStatusSerializer(_serialize_workspace(lib)).data)
+
+    # DELETE — idempotent: a missing workspace returns 204.
+    try:
+        lib = Library.nodes.get(workspace_id=workspace_id)
+    except Library.DoesNotExist:
+        return Response(status=status.HTTP_204_NO_CONTENT)
+
+    library_uid = lib.uid
+    library_name = lib.name
+
+    # Step 1-4: delete chunks, items, collections, then the library itself.
+    # We collect Item s3_keys first so the caller can clean up S3
+    # asynchronously (a future enhancement — for now, the keys are logged).
+    s3_rows, _ = db.cypher_query(
+        "MATCH (l:Library {workspace_id: $wsid})-[:CONTAINS]->(:Collection)"
+        "-[:CONTAINS]->(i:Item) RETURN i.uid, i.s3_key",
+        {"wsid": workspace_id},
+    )
+    item_s3_keys = [(r[0], r[1]) for r in s3_rows if r[1]]
+
+    db.cypher_query(
+        """
+        MATCH (l:Library {workspace_id: $wsid})-[:CONTAINS]->(:Collection)
+        -[:CONTAINS]->(i:Item)-[:HAS_CHUNK]->(c:Chunk)
+        DETACH DELETE c
+        """,
+        {"wsid": workspace_id},
+    )
+    db.cypher_query(
+        """
+        MATCH (l:Library {workspace_id: $wsid})-[:CONTAINS]->(:Collection)
+        -[:CONTAINS]->(i:Item)-[:HAS_IMAGE]->(img:Image)
+        OPTIONAL MATCH (img)-[:HAS_EMBEDDING]->(emb:ImageEmbedding)
+        DETACH DELETE img, emb
+        """,
+        {"wsid": workspace_id},
+    )
+    db.cypher_query(
+        """
+        MATCH (l:Library {workspace_id: $wsid})-[:CONTAINS]->(:Collection)
+        -[:CONTAINS]->(i:Item)
+        DETACH DELETE i
+        """,
+        {"wsid": workspace_id},
+    )
+    db.cypher_query(
+        """
+        MATCH (l:Library {workspace_id: $wsid})-[:CONTAINS]->(col:Collection)
+        DETACH DELETE col
+        """,
+        {"wsid": workspace_id},
+    )
+    db.cypher_query(
+        "MATCH (l:Library {workspace_id: $wsid}) DETACH DELETE l",
+        {"wsid": workspace_id},
+    )
+
+    # Step 5: orphan Concept garbage collection.
+    orphan_result, _ = db.cypher_query(
+        """
+        MATCH (con:Concept)
+        WHERE NOT (con)<-[:REFERENCES]-() AND NOT (con)<-[:MENTIONS]-()
+              AND NOT (con)<-[:DEPICTS]-()
+        WITH con
+        DETACH DELETE con
+        RETURN count(con) AS deleted
+        """
+    )
+    orphans_deleted = orphan_result[0][0] if orphan_result else 0
+
+    logger.info(
+        "Workspace deleted workspace_id=%s library_uid=%s name=%s "
+        "items=%d orphans_deleted=%d",
+        workspace_id, library_uid, library_name,
+        len(item_s3_keys), orphans_deleted,
+    )
+
+    return Response(status=status.HTTP_204_NO_CONTENT)

mnemosyne/library/content_types.py

@@ -210,6 +210,69 @@ LIBRARY_TYPE_DEFAULTS = {
             "4) Context clues about when and where this was taken or created."
         ),
     },
+    "business": {
+        "chunking_config": {
+            "strategy": "section_aware",
+            "chunk_size": 640,
+            "chunk_overlap": 96,
+            "respect_boundaries": ["section", "subsection", "list", "table"],
+        },
+        "embedding_instruction": (
+            "Represent this passage from a business document for retrieval. "
+            "Focus on value propositions, positioning, pricing, scope of work, "
+            "client outcomes, and commercial commitments."
+        ),
+        "reranker_instruction": (
+            "Re-rank passages from business documents based on commercial relevance. "
+            "Prioritize value framing, deliverables, client outcomes, and specific "
+            "pricing or scope language."
+        ),
+        "llm_context_prompt": (
+            "The following excerpts are from business documents (proposals, marketing, "
+            "sales, strategy). Interpret in commercial context. Distinguish positioning "
+            "claims from committed deliverables. Preserve numbers, scope language, and "
+            "client names exactly as written."
+        ),
+        "vision_prompt": (
+            "Analyze this image from a business document. Identify:\n"
+            "1) Image type (logo, chart, diagram, screenshot, photograph, table).\n"
+            "2) What it depicts — brand marks, data, organizational structure, products.\n"
+            "3) Any visible text — company names, figures, captions, headings.\n"
+            "4) The commercial purpose — positioning, pricing, capability demonstration."
+        ),
+    },
+    "finance": {
+        "chunking_config": {
+            "strategy": "section_aware",
+            "chunk_size": 512,
+            "chunk_overlap": 64,
+            "respect_boundaries": ["section", "table", "row", "paragraph"],
+        },
+        "embedding_instruction": (
+            "Represent this passage from a financial document for retrieval. "
+            "Focus on accounts, instruments, dates, amounts, balances, and "
+            "analytical commentary."
+        ),
+        "reranker_instruction": (
+            "Re-rank passages from financial documents based on relevance to the query. "
+            "Prioritize the matching account, instrument, time period, and figures."
+        ),
+        "llm_context_prompt": (
+            "The following excerpts are from financial documents (statements, tax "
+            "documents, market commentary, planning). Distinguish factual figures "
+            "(statements, transactions, balances) from opinion (forecasts, commentary). "
+            "Quote numbers, dates, and account identifiers exactly as they appear. "
+            "Do not infer, round, or fabricate financial figures. If a figure is not "
+            "present in the excerpts, say so explicitly."
+        ),
+        "vision_prompt": (
+            "Analyze this image from a financial document. Identify:\n"
+            "1) Image type (chart, table, statement scan, dashboard screenshot, receipt).\n"
+            "2) What it depicts — account, instrument, time period, data series.\n"
+            "3) Any visible text — figures, dates, account identifiers, labels.\n"
+            "4) Whether the data is factual (statement) or analytical (forecast/commentary)."
+        ),
+    },
 }
 
 
@@ -219,7 +282,7 @@ def get_library_type_config(library_type):
 
     Args:
         library_type: One of 'fiction', 'nonfiction', 'technical', 'music',
-                      'film', 'art', 'journal'
+                      'film', 'art', 'journal', 'business', 'finance'
 
     Returns:
         dict with keys: chunking_config, embedding_instruction,

mnemosyne/library/migrations/0001_initial.py

@@ -0,0 +1,45 @@
+# Generated by Django 5.2.13 on 2026-04-28 12:36
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    initial = True
+
+    dependencies = [
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name='IngestJob',
+            fields=[
+                ('id', models.CharField(max_length=64, primary_key=True, serialize=False)),
+                ('item_uid', models.CharField(blank=True, db_index=True, max_length=64)),
+                ('library_uid', models.CharField(db_index=True, max_length=64)),
+                ('celery_task_id', models.CharField(blank=True, max_length=255)),
+                ('status', models.CharField(choices=[('pending', 'Pending'), ('processing', 'Processing'), ('completed', 'Completed'), ('failed', 'Failed')], db_index=True, default='pending', max_length=20)),
+                ('progress', models.CharField(default='queued', max_length=50)),
+                ('error', models.TextField(blank=True, null=True)),
+                ('retry_count', models.PositiveIntegerField(default=0)),
+                ('chunks_created', models.PositiveIntegerField(default=0)),
+                ('concepts_extracted', models.PositiveIntegerField(default=0)),
+                ('embedding_model', models.CharField(blank=True, max_length=100)),
+                ('content_hash', models.CharField(blank=True, db_index=True, max_length=64)),
+                ('source', models.CharField(default='', max_length=50)),
+                ('source_ref', models.CharField(blank=True, db_index=True, max_length=200)),
+                ('s3_key', models.CharField(max_length=500)),
+                ('title', models.CharField(blank=True, max_length=500)),
+                ('file_type', models.CharField(blank=True, max_length=50)),
+                ('file_size', models.PositiveBigIntegerField(default=0)),
+                ('collection_uid', models.CharField(blank=True, max_length=64)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('started_at', models.DateTimeField(blank=True, null=True)),
+                ('completed_at', models.DateTimeField(blank=True, null=True)),
+            ],
+            options={
+                'ordering': ['-created_at'],
+                'indexes': [models.Index(fields=['status', '-created_at'], name='library_ing_status_9c95b2_idx'), models.Index(fields=['source', 'source_ref'], name='library_ing_source_a48684_idx')],
+            },
+        ),
+    ]

mnemosyne/library/models.py

@@ -1,11 +1,16 @@
 """
-Neo4j graph models for the Mnemosyne content library.
+Models for the Mnemosyne content library.
 
-All content data (libraries, collections, items, chunks, concepts, images)
-lives in Neo4j as a knowledge graph. These models use neomodel's StructuredNode
-OGM — they do NOT participate in Django's ORM or migrations.
+Most content (libraries, collections, items, chunks, concepts, images)
+lives in Neo4j as a knowledge graph via neomodel StructuredNode. These do
+NOT participate in Django's ORM or migrations.
+
+The IngestJob model at the bottom of this file is the exception: it tracks
+the lifecycle of asynchronous ingestion requests (file → embedding pipeline)
+in PostgreSQL via Django's ORM.
 """
 
+from django.db import models
 from neomodel import (
     ArrayProperty,
     DateTimeProperty,
@@ -50,8 +55,14 @@ class Library(StructuredNode):
     """
     Top-level container representing a content library.
 
-    Each library has a type (fiction, technical, music, film, art, journal)
-    that drives chunking strategy, embedding instructions, and LLM prompts.
+    Each library has a type (fiction, nonfiction, technical, music, film,
+    art, journal, business, finance) that drives chunking strategy,
+    embedding instructions, and LLM prompts.
+
+    A library may be either *global* (workspace_id is null — searchable
+    across the whole instance) or *workspace-scoped* (workspace_id set —
+    visible only to agents inside that Daedalus workspace). Scoping is
+    enforced structurally by every search query.
     """
 
     uid = UniqueIdProperty()
@@ -66,10 +77,16 @@ class Library(StructuredNode):
             "film": "Film",
             "art": "Art",
             "journal": "Journal",
+            "business": "Business",
+            "finance": "Finance",
         },
     )
     description = StringProperty(default="")
 
+    # Daedalus workspace UUID this library is scoped to. Null for global
+    # libraries. Unique-indexed so a workspace cannot have two libraries.
+    workspace_id = StringProperty(unique_index=True, required=False)
+
     # Content-type configuration
     chunking_config = JSONProperty(default={})
     embedding_instruction = StringProperty(default="")
@@ -270,3 +287,78 @@ class ImageEmbedding(StructuredNode):
 
     def __str__(self):
         return f"ImageEmbedding ({self.uid})"
+
+
+# --- Django ORM models (PostgreSQL) ---
+
+
+class IngestJob(models.Model):
+    """
+    Tracks the lifecycle of an asynchronous ingestion + embedding job.
+
+    Created when an external client (e.g. Daedalus) posts a file via the
+    REST ingest API. The Celery worker reads and updates this row as the
+    job moves through fetch / chunk / embed / graph stages.
+
+    Idempotency: a (library, source_ref, content_hash) triple uniquely
+    identifies a piece of content. A second POST with the same triple
+    returns the existing job; a POST with the same source_ref but a new
+    content_hash supersedes the prior Item.
+    """
+
+    STATUS_CHOICES = [
+        ("pending", "Pending"),
+        ("processing", "Processing"),
+        ("completed", "Completed"),
+        ("failed", "Failed"),
+    ]
+
+    id = models.CharField(max_length=64, primary_key=True)
+    item_uid = models.CharField(max_length=64, db_index=True, blank=True)
+    library_uid = models.CharField(max_length=64, db_index=True)
+    celery_task_id = models.CharField(max_length=255, blank=True)
+
+    status = models.CharField(
+        max_length=20,
+        choices=STATUS_CHOICES,
+        default="pending",
+        db_index=True,
+    )
+    progress = models.CharField(max_length=50, default="queued")
+    error = models.TextField(blank=True, null=True)
+    retry_count = models.PositiveIntegerField(default=0)
+
+    chunks_created = models.PositiveIntegerField(default=0)
+    concepts_extracted = models.PositiveIntegerField(default=0)
+    embedding_model = models.CharField(max_length=100, blank=True)
+
+    # The file's content hash (sha256). Used for idempotency: a second
+    # ingest with the same source_ref + same hash is a no-op; a second
+    # ingest with the same source_ref + different hash supersedes.
+    content_hash = models.CharField(max_length=64, db_index=True, blank=True)
+
+    # Where the file came from. For Daedalus: source="daedalus",
+    # source_ref="<workspace_id>/<file_id>".
+    source = models.CharField(max_length=50, default="")
+    source_ref = models.CharField(max_length=200, blank=True, db_index=True)
+    s3_key = models.CharField(max_length=500)
+
+    # Optional metadata carried forward to the Item node.
+    title = models.CharField(max_length=500, blank=True)
+    file_type = models.CharField(max_length=50, blank=True)
+    file_size = models.PositiveBigIntegerField(default=0)
+    collection_uid = models.CharField(max_length=64, blank=True)
+
+    created_at = models.DateTimeField(auto_now_add=True)
+    started_at = models.DateTimeField(null=True, blank=True)
+    completed_at = models.DateTimeField(null=True, blank=True)
+
+    class Meta:
+        ordering = ["-created_at"]
+        indexes = [
+            models.Index(fields=["status", "-created_at"]),
+            models.Index(fields=["source", "source_ref"]),
+        ]
+
+    def __str__(self):
+        return f"IngestJob {self.id} [{self.status}]"

mnemosyne/library/services/daedalus_s3.py

@@ -0,0 +1,70 @@
+"""
+Cross-bucket S3 helper for ingesting files from the Daedalus S3 bucket
+into Mnemosyne's own bucket.
+
+Daedalus uploads files to its own bucket (configured per Daedalus deployment)
+and posts an ingest request to Mnemosyne with the s3_key. This module fetches
+that file using read-only Daedalus credentials and writes it to Mnemosyne's
+bucket via the standard `default_storage` backend so the rest of the pipeline
+(parsing, chunking, embedding) works unchanged.
+"""
+
+import logging
+
+import boto3
+from django.conf import settings
+from django.core.files.base import ContentFile
+from django.core.files.storage import default_storage
+
+logger = logging.getLogger(__name__)
+
+
+def _daedalus_s3_client():
+    """Build a boto3 S3 client pointed at the Daedalus bucket."""
+    return boto3.client(
+        "s3",
+        endpoint_url=settings.DAEDALUS_S3_ENDPOINT_URL or None,
+        aws_access_key_id=settings.DAEDALUS_S3_ACCESS_KEY_ID,
+        aws_secret_access_key=settings.DAEDALUS_S3_SECRET_ACCESS_KEY,
+        region_name=settings.DAEDALUS_S3_REGION_NAME,
+        use_ssl=settings.DAEDALUS_S3_USE_SSL,
+        verify=settings.DAEDALUS_S3_VERIFY,
+    )
+
+
+def fetch_from_daedalus(daedalus_s3_key: str) -> bytes:
+    """
+    Read a file from the Daedalus S3 bucket.
+
+    :param daedalus_s3_key: Object key in the Daedalus bucket.
+    :returns: File bytes.
+    :raises: botocore exceptions on failure (caller decides retry).
+    """
+    client = _daedalus_s3_client()
+    bucket = settings.DAEDALUS_S3_BUCKET_NAME
+    logger.debug(
+        "Fetching from Daedalus S3 bucket=%s key=%s", bucket, daedalus_s3_key
+    )
+    response = client.get_object(Bucket=bucket, Key=daedalus_s3_key)
+    data = response["Body"].read()
+    logger.info(
+        "Fetched from Daedalus S3 bucket=%s key=%s size=%d",
+        bucket, daedalus_s3_key, len(data),
+    )
+    return data
+
+
+def copy_into_mnemosyne(data: bytes, mnemosyne_s3_key: str) -> str:
+    """
+    Write bytes into Mnemosyne's S3 bucket via the default_storage backend.
+
+    :param data: File bytes (already in memory).
+    :param mnemosyne_s3_key: Target object key in Mnemosyne's bucket.
+    :returns: The actual key written (may differ from requested if
+              `file_overwrite=False` and the key existed).
+    """
+    saved_key = default_storage.save(mnemosyne_s3_key, ContentFile(data))
+    logger.info(
+        "Wrote to Mnemosyne S3 key=%s size=%d", saved_key, len(data),
+    )
+    return saved_key

mnemosyne/library/services/search.py

@@ -26,15 +26,32 @@ from .fusion import ImageSearchResult, SearchCandidate, reciprocal_rank_fusion
 logger = logging.getLogger(__name__)
 
 
+# Workspace scoping clause appended to every search Cypher query.
+#
+# A request with workspace_id set returns ONLY that workspace's content.
+# A request with workspace_id null returns ONLY global content (libraries
+# with no workspace_id). There is no third mode.
+_WORKSPACE_SCOPE_CLAUSE = (
+    " AND ($workspace_id IS NULL AND lib.workspace_id IS NULL OR "
+    "lib.workspace_id = $workspace_id)"
+)
+
+
 @dataclass
 class SearchRequest:
-    """Parameters for a search query."""
+    """Parameters for a search query.
+
+    Scope is single-mode: a request is either workspace-scoped (workspace_id
+    set) or global (workspace_id is None). There is no parameter combination
+    that returns both workspace and global content in one call.
+    """
 
     query: str
     query_image: Optional[bytes] = None
     library_uid: Optional[str] = None
     library_type: Optional[str] = None
     collection_uid: Optional[str] = None
+    workspace_id: Optional[str] = None
     search_types: list[str] = field(
         default_factory=lambda: ["vector", "fulltext", "graph"]
     )
@@ -45,6 +62,18 @@ class SearchRequest:
     rerank: bool = True
     include_images: bool = True
 
+    def __post_init__(self):
+        # Normalize empty strings to None so "" doesn't slip through as
+        # truthy at the Cypher boundary.
+        if self.workspace_id == "":
+            self.workspace_id = None
+        if self.library_uid == "":
+            self.library_uid = None
+        if self.library_type == "":
+            self.library_type = None
+        if self.collection_uid == "":
+            self.collection_uid = None
+
 
 @dataclass
 class SearchResponse:
@@ -243,7 +272,8 @@ class SearchService:
         top_k = request.vector_top_k
 
         # Build Cypher with optional filtering
-        cypher = """
+        cypher = (
+            """
             CALL db.index.vector.queryNodes('chunk_embedding_index', $top_k, $query_vector)
             YIELD node AS chunk, score
             MATCH (item:Item)-[:HAS_CHUNK]->(chunk)
@@ -251,13 +281,17 @@ class SearchService:
             WHERE ($library_uid IS NULL OR lib.uid = $library_uid)
               AND ($library_type IS NULL OR lib.library_type = $library_type)
               AND ($collection_uid IS NULL OR col.uid = $collection_uid)
+            """
+            + _WORKSPACE_SCOPE_CLAUSE
+            + """
             RETURN chunk.uid AS chunk_uid, chunk.text_preview AS text_preview,
                    chunk.chunk_s3_key AS chunk_s3_key, chunk.chunk_index AS chunk_index,
                    item.uid AS item_uid, item.title AS item_title,
                    lib.library_type AS library_type, score
             ORDER BY score DESC
             LIMIT $top_k
-        """
+            """
+        )
 
         params = {
             "top_k": top_k,
@@ -265,6 +299,7 @@ class SearchService:
             "library_uid": request.library_uid,
             "library_type": request.library_type,
             "collection_uid": request.collection_uid,
+            "workspace_id": request.workspace_id,
         }
 
         try:
@@ -348,7 +383,8 @@ class SearchService:
         candidates: dict[str, SearchCandidate],
     ):
         """Search chunk_text_fulltext index and add to candidates dict."""
-        cypher = """
+        cypher = (
+            """
             CALL db.index.fulltext.queryNodes('chunk_text_fulltext', $query)
             YIELD node AS chunk, score
             MATCH (item:Item)-[:HAS_CHUNK]->(chunk)
@@ -356,13 +392,17 @@ class SearchService:
             WHERE ($library_uid IS NULL OR lib.uid = $library_uid)
               AND ($library_type IS NULL OR lib.library_type = $library_type)
               AND ($collection_uid IS NULL OR col.uid = $collection_uid)
+            """
+            + _WORKSPACE_SCOPE_CLAUSE
+            + """
             RETURN chunk.uid AS chunk_uid, chunk.text_preview AS text_preview,
                    chunk.chunk_s3_key AS chunk_s3_key, chunk.chunk_index AS chunk_index,
                    item.uid AS item_uid, item.title AS item_title,
                    lib.library_type AS library_type, score
             ORDER BY score DESC
             LIMIT $top_k
-        """
+            """
+        )
 
         params = {
             "query": request.query,
@@ -370,6 +410,7 @@ class SearchService:
             "library_uid": request.library_uid,
             "library_type": request.library_type,
             "collection_uid": request.collection_uid,
+            "workspace_id": request.workspace_id,
         }
 
         try:
@@ -402,7 +443,8 @@ class SearchService:
         candidates: dict[str, SearchCandidate],
     ):
         """Search concept_name_fulltext and traverse to chunks."""
-        cypher = """
+        cypher = (
+            """
             CALL db.index.fulltext.queryNodes('concept_name_fulltext', $query)
             YIELD node AS concept, score AS concept_score
             MATCH (chunk:Chunk)-[:MENTIONS]->(concept)
@@ -410,6 +452,9 @@ class SearchService:
             MATCH (lib:Library)-[:CONTAINS]->(:Collection)-[:CONTAINS]->(item)
             WHERE ($library_uid IS NULL OR lib.uid = $library_uid)
               AND ($library_type IS NULL OR lib.library_type = $library_type)
+            """
+            + _WORKSPACE_SCOPE_CLAUSE
+            + """
             RETURN chunk.uid AS chunk_uid, chunk.text_preview AS text_preview,
                    chunk.chunk_s3_key AS chunk_s3_key, chunk.chunk_index AS chunk_index,
                    item.uid AS item_uid, item.title AS item_title,
@@ -417,13 +462,15 @@ class SearchService:
                    concept_score * 0.8 AS score
             ORDER BY score DESC
             LIMIT $top_k
-        """
+            """
+        )
 
         params = {
             "query": request.query,
             "top_k": top_k,
             "library_uid": request.library_uid,
             "library_type": request.library_type,
+            "workspace_id": request.workspace_id,
         }
 
         try:
@@ -465,7 +512,8 @@ class SearchService:
         """
         start = time.time()
 
-        cypher = """
+        cypher = (
+            """
             CALL db.index.fulltext.queryNodes('concept_name_fulltext', $query)
             YIELD node AS concept, score AS concept_score
             WITH concept, concept_score
@@ -476,6 +524,9 @@ class SearchService:
             MATCH (lib:Library)-[:CONTAINS]->(:Collection)-[:CONTAINS]->(item)
             WHERE ($library_uid IS NULL OR lib.uid = $library_uid)
               AND ($library_type IS NULL OR lib.library_type = $library_type)
+            """
+            + _WORKSPACE_SCOPE_CLAUSE
+            + """
             WITH chunk, item, lib,
                  max(concept_score) AS score,
                  collect(DISTINCT concept.name)[..5] AS concept_names
@@ -486,13 +537,15 @@ class SearchService:
                    score, concept_names
             ORDER BY score DESC
             LIMIT $limit
-        """
+            """
+        )
 
         params = {
             "query": request.query,
             "limit": request.fulltext_top_k,
             "library_uid": request.library_uid,
             "library_type": request.library_type,
+            "workspace_id": request.workspace_id,
         }
 
         try:
@@ -550,7 +603,8 @@ class SearchService:
         """
         start = time.time()
 
-        cypher = """
+        cypher = (
+            """
             CALL db.index.vector.queryNodes('image_embedding_index', $top_k, $query_vector)
             YIELD node AS emb_node, score
             MATCH (img:Image)-[:HAS_EMBEDDING]->(emb_node)
@@ -558,19 +612,24 @@ class SearchService:
             MATCH (lib:Library)-[:CONTAINS]->(:Collection)-[:CONTAINS]->(item)
             WHERE ($library_uid IS NULL OR lib.uid = $library_uid)
               AND ($library_type IS NULL OR lib.library_type = $library_type)
+            """
+            + _WORKSPACE_SCOPE_CLAUSE
+            + """
             RETURN img.uid AS image_uid, img.image_type AS image_type,
                    img.description AS description, img.s3_key AS s3_key,
                    item.uid AS item_uid, item.title AS item_title,
                    score
             ORDER BY score DESC
             LIMIT 10
-        """
+            """
+        )
 
         params = {
             "top_k": 10,
             "query_vector": query_vector,
             "library_uid": request.library_uid,
             "library_type": request.library_type,
+            "workspace_id": request.workspace_id,
         }
 
         try:

mnemosyne/library/tasks.py

@@ -10,6 +10,7 @@ import logging
 
 from celery import shared_task
 from django.core.cache import cache
+from neomodel import db
 
 logger = logging.getLogger(__name__)
 
@@ -280,3 +281,212 @@ def _resolve_user(user_id: int = None):
         return User.objects.get(pk=user_id)
     except Exception:
         return None
+
+
+# ---------------------------------------------------------------------------
+# Ingest task (Daedalus integration)
+# ---------------------------------------------------------------------------
+
+
+@shared_task(
+    name="library.tasks.ingest_from_daedalus",
+    bind=True,
+    queue="embedding",
+    max_retries=3,
+    default_retry_delay=60,
+    acks_late=True,
+)
+def ingest_from_daedalus(self, job_id: str):
+    """
+    Process a single IngestJob: fetch from Daedalus S3 → create Item →
+    run embedding pipeline → mark complete.
+
+    Idempotent on (library_uid, source_ref, content_hash) — handled in the
+    REST view that creates the IngestJob, so by the time this task runs the
+    job either represents new content or a content_hash-changed re-ingest.
+
+    For a content_hash-changed re-ingest, the prior Item with the same
+    source_ref is deleted before the new one is processed (ensures no
+    stale chunks linger).
+    """
+    from datetime import datetime, timezone
+
+    from library.models import IngestJob, Item, Library
+    from library.services.daedalus_s3 import (
+        copy_into_mnemosyne,
+        fetch_from_daedalus,
+    )
+    from library.services.pipeline import EmbeddingPipeline
+
+    logger.info(
+        "Task ingest_from_daedalus starting job_id=%s task_id=%s",
+        job_id, self.request.id,
+    )
+
+    try:
+        job = IngestJob.objects.get(pk=job_id)
+    except IngestJob.DoesNotExist:
+        logger.error("IngestJob not found job_id=%s", job_id)
+        return {"success": False, "error": "job_not_found"}
+
+    job.status = "processing"
+    job.progress = "fetching"
+    job.started_at = datetime.now(timezone.utc)
+    job.celery_task_id = self.request.id
+    job.save(update_fields=["status", "progress", "started_at", "celery_task_id"])
+
+    try:
+        # --- 1. Resolve target Library ---
+        try:
+            lib = Library.nodes.get(uid=job.library_uid)
+        except Library.DoesNotExist:
+            raise RuntimeError(f"Library not found: {job.library_uid}")
+
+        # --- 2. Supersede prior Item with same source_ref but different hash ---
+        prior_item_uid = None
+        if job.source_ref:
+            rows, _ = db.cypher_query(
+                """
+                MATCH (l:Library {uid: $library_uid})-[:CONTAINS]->(:Collection)
+                -[:CONTAINS]->(i:Item)
+                WHERE i.metadata IS NOT NULL
+                  AND i.metadata CONTAINS $source_ref_marker
+                RETURN i.uid LIMIT 1
+                """,
+                {
+                    "library_uid": lib.uid,
+                    "source_ref_marker": f'"source_ref": "{job.source_ref}"',
+                },
+            )
+            if rows:
+                prior_item_uid = rows[0][0]
+                logger.info(
+                    "Superseding prior Item job_id=%s prior_item_uid=%s",
+                    job_id, prior_item_uid,
+                )
+                _delete_item_and_chunks(prior_item_uid)
+
+        # --- 3. Fetch from Daedalus, copy into Mnemosyne bucket ---
+        job.progress = "copying"
+        job.save(update_fields=["progress"])
+
+        data = fetch_from_daedalus(job.s3_key)
+
+        # --- 4. Create Item node ---
+        ext = (job.file_type or "bin").lstrip(".").lower() or "bin"
+        item = Item(
+            title=job.title,
+            file_type=ext,
+            file_size=len(data),
+            content_hash=job.content_hash,
+            embedding_status="pending",
+            metadata={
+                "source": job.source,
+                "source_ref": job.source_ref,
+            },
+        )
+        item.save()
+
+        mnemosyne_s3_key = f"items/{item.uid}/original.{ext}"
+        copy_into_mnemosyne(data, mnemosyne_s3_key)
+        item.s3_key = mnemosyne_s3_key
+        item.save()
+
+        # --- 5. Connect to library/collection ---
+        col = _resolve_or_create_default_collection(lib, job.collection_uid)
+        col.items.connect(item)
+
+        job.item_uid = item.uid
+        job.save(update_fields=["item_uid"])
+
+        # --- 6. Run the embedding pipeline ---
+        job.progress = "embedding"
+        job.save(update_fields=["progress"])
+
+        def progress_cb(percent, message):
+            _update_progress(self, percent, message)
+
+        pipeline = EmbeddingPipeline(user=None)
+        result = pipeline.process_item(item.uid, progress_callback=progress_cb)
+
+        # --- 7. Mark complete ---
+        job.status = "completed"
+        job.progress = "done"
+        job.chunks_created = result.get("chunks_created", 0)
+        job.concepts_extracted = result.get("concepts_extracted", 0)
+        job.embedding_model = result.get("embedding_model", "")
+        job.completed_at = datetime.now(timezone.utc)
+        job.save()
+
+        logger.info(
+            "Task ingest_from_daedalus completed job_id=%s item_uid=%s "
+            "chunks=%d concepts=%d",
+            job_id, item.uid, job.chunks_created, job.concepts_extracted,
+        )
+        return {
+            "success": True,
+            "job_id": job_id,
+            "item_uid": item.uid,
+            **result,
+        }
+
+    except Exception as exc:
+        logger.error(
+            "Task ingest_from_daedalus failed job_id=%s: %s",
+            job_id, exc, exc_info=True,
+        )
+        if self.request.retries < self.max_retries:
+            job.retry_count = self.request.retries + 1
+            job.save(update_fields=["retry_count"])
+            raise self.retry(exc=exc)
+
+        job.status = "failed"
+        job.error = str(exc)
+        job.completed_at = datetime.now(timezone.utc)
+        job.save(update_fields=["status", "error", "completed_at"])
+        return {"success": False, "job_id": job_id, "error": str(exc)}
+
+
+def _delete_item_and_chunks(item_uid: str):
+    """Delete an Item, its chunks, and its images. Concept GC is workspace-delete only."""
+    db.cypher_query(
+        """
+        MATCH (i:Item {uid: $uid})
+        OPTIONAL MATCH (i)-[:HAS_CHUNK]->(c:Chunk)
+        OPTIONAL MATCH (i)-[:HAS_IMAGE]->(img:Image)
+        OPTIONAL MATCH (img)-[:HAS_EMBEDDING]->(emb:ImageEmbedding)
+        DETACH DELETE c, img, emb, i
+        """,
+        {"uid": item_uid},
+    )
+
+
+def _resolve_or_create_default_collection(lib, collection_uid: str = ""):
+    """
+    Find or create the default Collection for a Library.
+
+    Daedalus integration creates one Collection per Library, named "default".
+    Explicit collection_uid is honored if provided.
+    """
+    from library.models import Collection
+
+    if collection_uid:
+        try:
+            return Collection.nodes.get(uid=collection_uid)
+        except Collection.DoesNotExist:
+            pass
+
+    # Look for an existing "default" collection in this library
+    rows, _ = db.cypher_query(
+        "MATCH (l:Library {uid: $library_uid})-[:CONTAINS]->(c:Collection {name: 'default'}) "
+        "RETURN c.uid LIMIT 1",
+        {"library_uid": lib.uid},
+    )
+    if rows:
+        return Collection.nodes.get(uid=rows[0][0])
+
+    col = Collection(name="default", description="Default collection")
+    col.save()
+    lib.collections.connect(col)
+    col.library.connect(lib)
+    return col

mnemosyne/library/tests/test_content_types.py

@@ -13,7 +13,17 @@ from library.content_types import LIBRARY_TYPE_DEFAULTS, get_library_type_config
 class LibraryTypeDefaultsTests(TestCase):
     """Tests for the LIBRARY_TYPE_DEFAULTS registry."""
 
-    EXPECTED_TYPES = {"fiction", "nonfiction", "technical", "music", "film", "art", "journal"}
+    EXPECTED_TYPES = {
+        "fiction",
+        "nonfiction",
+        "technical",
+        "music",
+        "film",
+        "art",
+        "journal",
+        "business",
+        "finance",
+    }
 
     def test_all_expected_types_present(self):
         for lib_type in self.EXPECTED_TYPES:
@@ -105,6 +115,16 @@ class VisionPromptTests(TestCase):
         prompt = config["vision_prompt"].lower()
         self.assertIn("historical", prompt)
 
+    def test_business_vision_prompt_mentions_logo_or_chart(self):
+        config = get_library_type_config("business")
+        prompt = config["vision_prompt"].lower()
+        self.assertTrue("logo" in prompt or "chart" in prompt)
+
+    def test_finance_llm_context_forbids_fabrication(self):
+        config = get_library_type_config("finance")
+        prompt = config["llm_context_prompt"].lower()
+        self.assertIn("fabricate", prompt)
+
 
 class GetLibraryTypeConfigTests(TestCase):
     """Tests for the get_library_type_config helper."""

mnemosyne/library/tests/test_search_scoping.py

@@ -0,0 +1,71 @@
+"""
+Tests for workspace scoping in SearchRequest and the Cypher scope clause.
+
+These exercise the dataclass-level normalization and the construction
+of Cypher parameter dicts. The actual Cypher execution against Neo4j
+is validated by the manual end-to-end test plan.
+"""
+
+from django.test import TestCase
+
+from library.services.search import _WORKSPACE_SCOPE_CLAUSE, SearchRequest
+
+
+class SearchRequestScopingTests(TestCase):
+    """SearchRequest workspace_id behavior."""
+
+    def test_default_workspace_id_is_none(self):
+        req = SearchRequest(query="hello")
+        self.assertIsNone(req.workspace_id)
+
+    def test_explicit_workspace_id_preserved(self):
+        req = SearchRequest(query="hello", workspace_id="ws_abc")
+        self.assertEqual(req.workspace_id, "ws_abc")
+
+    def test_empty_string_workspace_id_normalized_to_none(self):
+        """Empty strings must NOT slip through as a truthy filter at the Cypher boundary."""
+        req = SearchRequest(query="hello", workspace_id="")
+        self.assertIsNone(req.workspace_id)
+
+    def test_empty_string_library_uid_normalized_to_none(self):
+        req = SearchRequest(query="hello", library_uid="")
+        self.assertIsNone(req.library_uid)
+
+    def test_empty_string_library_type_normalized_to_none(self):
+        req = SearchRequest(query="hello", library_type="")
+        self.assertIsNone(req.library_type)
+
+    def test_empty_string_collection_uid_normalized_to_none(self):
+        req = SearchRequest(query="hello", collection_uid="")
+        self.assertIsNone(req.collection_uid)
+
+
+class WorkspaceScopeClauseTests(TestCase):
+    """Sanity checks on the Cypher snippet itself.
+
+    The clause must produce two distinct, non-overlapping result sets:
+    1. workspace_id IS NULL → only global libraries (lib.workspace_id IS NULL)
+    2. workspace_id = X      → only libraries with workspace_id = X
+
+    A "leaks both" bug would be a Cypher OR that fails to bracket properly.
+    Verifying the literal string here is a cheap regression guard against
+    refactors that accidentally change the operator precedence.
+    """
+
+    def test_clause_references_lib_workspace_id(self):
+        self.assertIn("lib.workspace_id", _WORKSPACE_SCOPE_CLAUSE)
+
+    def test_clause_references_workspace_id_param(self):
+        self.assertIn("$workspace_id", _WORKSPACE_SCOPE_CLAUSE)
+
+    def test_clause_handles_both_modes(self):
+        """Both 'IS NULL' and '=' branches must be present."""
+        self.assertIn("IS NULL", _WORKSPACE_SCOPE_CLAUSE)
+        self.assertIn("=", _WORKSPACE_SCOPE_CLAUSE)
+
+    def test_clause_starts_with_AND_so_it_appends_safely(self):
+        """The clause is appended to existing WHERE filters."""
+        self.assertTrue(
+            _WORKSPACE_SCOPE_CLAUSE.lstrip().startswith("AND"),
+            f"Clause must start with AND: {_WORKSPACE_SCOPE_CLAUSE!r}",
+        )

mnemosyne/library/tests/test_workspaces_api.py

@@ -0,0 +1,210 @@
+"""
+Tests for workspace and ingest REST endpoints.
+
+These exercise serializer validation, idempotency rules, and Django ORM
+behavior for IngestJob. The Cypher-touching paths (Library node CRUD,
+search scoping) require Neo4j and are validated by the manual end-to-end
+test plan, not these unit tests.
+"""
+
+from django.test import TestCase
+from rest_framework.test import APIClient
+
+from library.api.serializers import (
+    IngestRequestSerializer,
+    WorkspaceCreateSerializer,
+)
+from library.models import IngestJob
+
+
+class WorkspaceCreateSerializerTests(TestCase):
+    """Validation rules for the create-workspace payload."""
+
+    def test_minimal_payload_validates(self):
+        s = WorkspaceCreateSerializer(
+            data={
+                "workspace_id": "ws_abc",
+                "name": "My Workspace",
+                "library_type": "technical",
+            }
+        )
+        self.assertTrue(s.is_valid(), s.errors)
+
+    def test_business_type_accepted(self):
+        s = WorkspaceCreateSerializer(
+            data={
+                "workspace_id": "ws_abc",
+                "name": "Sales",
+                "library_type": "business",
+            }
+        )
+        self.assertTrue(s.is_valid(), s.errors)
+
+    def test_finance_type_accepted(self):
+        s = WorkspaceCreateSerializer(
+            data={
+                "workspace_id": "ws_abc",
+                "name": "Money",
+                "library_type": "finance",
+            }
+        )
+        self.assertTrue(s.is_valid(), s.errors)
+
+    def test_unknown_type_rejected(self):
+        s = WorkspaceCreateSerializer(
+            data={
+                "workspace_id": "ws_abc",
+                "name": "x",
+                "library_type": "miscellaneous",
+            }
+        )
+        self.assertFalse(s.is_valid())
+        self.assertIn("library_type", s.errors)
+
+    def test_workspace_id_required(self):
+        s = WorkspaceCreateSerializer(
+            data={"name": "x", "library_type": "technical"}
+        )
+        self.assertFalse(s.is_valid())
+        self.assertIn("workspace_id", s.errors)
+
+
+class IngestRequestSerializerTests(TestCase):
+    """Validation rules for the ingest payload."""
+
+    BASE_PAYLOAD = {
+        "s3_key": "workspaces/ws/files/f/x.pdf",
+        "title": "Q4 Report",
+        "content_hash": "a" * 64,
+    }
+
+    def test_workspace_id_only_validates(self):
+        s = IngestRequestSerializer(
+            data={**self.BASE_PAYLOAD, "workspace_id": "ws_abc"}
+        )
+        self.assertTrue(s.is_valid(), s.errors)
+
+    def test_library_uid_only_validates(self):
+        s = IngestRequestSerializer(
+            data={**self.BASE_PAYLOAD, "library_uid": "lib_xyz"}
+        )
+        self.assertTrue(s.is_valid(), s.errors)
+
+    def test_neither_workspace_nor_library_rejected(self):
+        s = IngestRequestSerializer(data=self.BASE_PAYLOAD)
+        self.assertFalse(s.is_valid())
+        # ValidationError on the whole payload, not a specific field
+        self.assertIn("non_field_errors", s.errors)
+
+    def test_content_hash_required(self):
+        s = IngestRequestSerializer(
+            data={
+                "s3_key": "x",
+                "title": "y",
+                "workspace_id": "ws_abc",
+            }
+        )
+        self.assertFalse(s.is_valid())
+        self.assertIn("content_hash", s.errors)
+
+
+class IngestJobModelTests(TestCase):
+    """IngestJob persists and queries correctly."""
+
+    def test_create_with_minimal_fields(self):
+        job = IngestJob.objects.create(
+            id="job_test1",
+            library_uid="lib_xyz",
+            s3_key="x.pdf",
+        )
+        self.assertEqual(job.status, "pending")
+        self.assertEqual(job.progress, "queued")
+        self.assertEqual(job.retry_count, 0)
+        self.assertEqual(job.chunks_created, 0)
+
+    def test_idempotency_query_pattern(self):
+        """The idempotency query in ingest_create uses (library, source_ref, hash)."""
+        IngestJob.objects.create(
+            id="job_a",
+            library_uid="lib_xyz",
+            source_ref="ws_a/file_1",
+            content_hash="h1",
+            s3_key="a.pdf",
+            status="completed",
+        )
+        IngestJob.objects.create(
+            id="job_b",
+            library_uid="lib_xyz",
+            source_ref="ws_a/file_1",
+            content_hash="h2",  # different hash — supersedes
+            s3_key="a.pdf",
+            status="completed",
+        )
+
+        # Same library + source_ref + h1 → finds job_a
+        match = IngestJob.objects.filter(
+            library_uid="lib_xyz",
+            source_ref="ws_a/file_1",
+            content_hash="h1",
+        ).first()
+        self.assertIsNotNone(match)
+        self.assertEqual(match.id, "job_a")
+
+        # h2 → finds job_b
+        match = IngestJob.objects.filter(
+            library_uid="lib_xyz",
+            source_ref="ws_a/file_1",
+            content_hash="h2",
+        ).first()
+        self.assertEqual(match.id, "job_b")
+
+        # Different source_ref → no match
+        match = IngestJob.objects.filter(
+            library_uid="lib_xyz",
+            source_ref="ws_a/file_2",
+            content_hash="h1",
+        ).first()
+        self.assertIsNone(match)
+
+
+class IngestEndpointAuthTests(TestCase):
+    """Auth boundary on the ingest endpoint (matches the existing convention)."""
+
+    def setUp(self):
+        self.client = APIClient()
+
+    def test_ingest_requires_auth(self):
+        response = self.client.post(
+            "/library/api/ingest/",
+            {
+                "s3_key": "x",
+                "title": "y",
+                "workspace_id": "ws_abc",
+                "content_hash": "a" * 64,
+            },
+            format="json",
+        )
+        self.assertIn(response.status_code, [401, 403])
+
+
+class WorkspaceEndpointAuthTests(TestCase):
+    """Auth on workspace endpoints."""
+
+    def setUp(self):
+        self.client = APIClient()
+
+    def test_workspace_create_requires_auth(self):
+        response = self.client.post(
+            "/library/api/workspaces/",
+            {"workspace_id": "ws_a", "name": "x", "library_type": "technical"},
+            format="json",
+        )
+        self.assertIn(response.status_code, [401, 403])
+
+    def test_workspace_get_requires_auth(self):
+        response = self.client.get("/library/api/workspaces/ws_a/")
+        self.assertIn(response.status_code, [401, 403])
+
+    def test_workspace_delete_requires_auth(self):
+        response = self.client.delete("/library/api/workspaces/ws_a/")
+        self.assertIn(response.status_code, [401, 403])

mnemosyne/mcp_server/admin.py

@@ -16,14 +16,23 @@ class MCPTokenAdmin(admin.ModelAdmin):
     ]
     list_filter = ["is_active"]
     search_fields = ["name", "user__email", "user__username"]
-    readonly_fields = ["token", "last_used_at", "created_at", "updated_at"]
+    readonly_fields = ["token_hash", "last_used_at", "created_at", "updated_at"]
     fieldsets = (
         (None, {"fields": ("user", "name", "is_active")}),
         ("Restrictions", {"fields": ("allowed_tools", "expires_at")}),
-        ("Token (shown once at creation)", {"fields": ("token",)}),
+        (
+            "Token (hashed at rest — plaintext is shown only once at creation)",
+            {"fields": ("token_hash",)},
+        ),
         ("Audit", {"fields": ("last_used_at", "created_at", "updated_at")}),
     )
 
     @admin.display(description="Token")
     def masked_token(self, obj):
         return obj.get_masked_token()
+
+    def has_add_permission(self, request):
+        # Tokens must be created via the dashboard or management command
+        # so the plaintext can be surfaced to the user. Adding via admin
+        # would persist a hash with no plaintext ever shown.
+        return False

mnemosyne/mcp_server/auth.py

@@ -12,7 +12,7 @@ from fastmcp.server.dependencies import get_http_request
 from fastmcp.server.middleware import Middleware, MiddlewareContext
 
 from .metrics import mcp_auth_failures_total
-from .models import MCPToken
+from .models import MCPToken, hash_token
 
 logger = logging.getLogger(__name__)
 
@@ -25,9 +25,17 @@ class MCPAuthError(Exception):
 
 
 def resolve_mcp_user(token_string: str):
-    """Resolve a bearer token to (user, MCPToken). Raises MCPAuthError on any failure."""
+    """Resolve a bearer token to (user, MCPToken). Raises MCPAuthError on any failure.
+
+    Hashes the incoming bearer and looks up by the hash — plaintext is never
+    stored or compared directly.
+    """
     try:
-        token = MCPToken.objects.select_related("user").get(token=token_string)
+        token = (
+            MCPToken.objects
+            .select_related("user")
+            .get(token_hash=hash_token(token_string))
+        )
     except MCPToken.DoesNotExist:
         raise MCPAuthError("Invalid MCP token.")
 

mnemosyne/mcp_server/forms.py

@@ -0,0 +1,85 @@
+"""Forms for the MCP token self-service dashboard."""
+
+from __future__ import annotations
+
+import asyncio
+import functools
+
+from django import forms
+
+from .models import MCPToken
+
+
+@functools.lru_cache(maxsize=1)
+def _registered_tool_names() -> list[str]:
+    """Pull the list of registered MCP tool names from the FastMCP instance.
+
+    Cached at module level — the tool registry is fixed at server startup.
+    Importing here (rather than at module import) avoids circulars when the
+    server module imports the dashboard pieces transitively.
+    """
+    from .server import mcp
+
+    tools = asyncio.run(mcp.get_tools())
+    return sorted(tools.keys())
+
+
+def _tool_choices() -> list[tuple[str, str]]:
+    return [(name, name) for name in _registered_tool_names()]
+
+
+class MCPTokenCreateForm(forms.Form):
+    """Generate a new bearer token. The token value itself is server-generated."""
+
+    name = forms.CharField(
+        max_length=100,
+        widget=forms.TextInput(attrs={
+            "class": "input input-bordered w-full",
+            "placeholder": "e.g. Claude Desktop, CI script",
+        }),
+    )
+    expires_at = forms.DateTimeField(
+        required=False,
+        widget=forms.DateTimeInput(attrs={
+            "class": "input input-bordered w-full",
+            "type": "datetime-local",
+        }),
+        help_text="Leave blank for no expiry.",
+    )
+    allowed_tools = forms.MultipleChoiceField(
+        required=False,
+        widget=forms.CheckboxSelectMultiple(attrs={"class": "checkbox checkbox-sm"}),
+        help_text="Leave all unchecked to permit every tool.",
+    )
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.fields["allowed_tools"].choices = _tool_choices()
+
+
+class MCPTokenEditForm(forms.ModelForm):
+    """Edit token metadata. The hashed token itself cannot be edited."""
+
+    allowed_tools = forms.MultipleChoiceField(
+        required=False,
+        widget=forms.CheckboxSelectMultiple(attrs={"class": "checkbox checkbox-sm"}),
+        help_text="Leave all unchecked to permit every tool.",
+    )
+
+    class Meta:
+        model = MCPToken
+        fields = ["name", "is_active", "expires_at", "allowed_tools"]
+        widgets = {
+            "name": forms.TextInput(attrs={"class": "input input-bordered w-full"}),
+            "is_active": forms.CheckboxInput(attrs={"class": "toggle toggle-primary"}),
+            "expires_at": forms.DateTimeInput(attrs={
+                "class": "input input-bordered w-full",
+                "type": "datetime-local",
+            }),
+        }
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.fields["allowed_tools"].choices = _tool_choices()
+        if self.instance and self.instance.pk:
+            self.fields["allowed_tools"].initial = self.instance.allowed_tools or []

mnemosyne/mcp_server/management/commands/create_mcp_token.py

@@ -57,7 +57,7 @@ class Command(BaseCommand):
                 raise CommandError("--expires-days must be at least 1.")
             expires_at = timezone.now() + timedelta(days=options["expires_days"])
 
-        token = MCPToken.objects.create(
+        token, plaintext = MCPToken.objects.create_token(
             user=user,
             name=options["name"],
             allowed_tools=allowed_tools,
@@ -73,5 +73,5 @@ class Command(BaseCommand):
             self.stdout.write("  Tools:   (all)")
         if expires_at:
             self.stdout.write(f"  Expires: {expires_at.isoformat()}")
-        self.stdout.write(self.style.WARNING("  Token (shown once):"))
-        self.stdout.write(f"  {token.token}")
+        self.stdout.write(self.style.WARNING("  Token (shown once — store it now):"))
+        self.stdout.write(f"  {plaintext}")

mnemosyne/mcp_server/migrations/0002_hash_token.py

@@ -0,0 +1,43 @@
+"""Hash MCPToken values at rest.
+
+Renames ``token`` → ``token_hash`` and rewrites any pre-existing plaintext
+values into SHA-256 hex digests in-place. Forward-only: hashing is one-way,
+so no reverse migration is provided.
+
+Existing tokens issued before this migration keep working only because
+``resolve_mcp_user`` hashes the incoming bearer before lookup; the original
+plaintext the client holds still hashes to what we just wrote.
+"""
+
+import hashlib
+
+from django.db import migrations, models
+
+
+def hash_existing_tokens(apps, schema_editor):
+    MCPToken = apps.get_model("mcp_server", "MCPToken")
+    for token in MCPToken.objects.all():
+        plaintext = token.token_hash  # post-rename, still holds original plaintext
+        token.token_hash = hashlib.sha256(plaintext.encode("utf-8")).hexdigest()
+        token.save(update_fields=["token_hash"])
+
+
+def noop_reverse(apps, schema_editor):
+    # Cannot reverse a hash. Leaving as no-op so the schema can be rolled
+    # back, but operators must understand any hashed rows are unrecoverable.
+    pass
+
+
+class Migration(migrations.Migration):
+    dependencies = [
+        ("mcp_server", "0001_initial"),
+    ]
+
+    operations = [
+        migrations.RenameField(
+            model_name="mcptoken",
+            old_name="token",
+            new_name="token_hash",
+        ),
+        migrations.RunPython(hash_existing_tokens, noop_reverse),
+    ]

mnemosyne/mcp_server/models.py

@@ -1,3 +1,4 @@
+import hashlib
 import secrets
 
 from django.conf import settings
@@ -5,15 +6,44 @@ from django.db import models
 from django.utils import timezone
 
 
+def hash_token(plaintext: str) -> str:
+    """SHA-256 hex digest of an MCP bearer token. 64 chars."""
+    return hashlib.sha256(plaintext.encode("utf-8")).hexdigest()
+
+
+class MCPTokenManager(models.Manager):
+    def create_token(self, *, user, name, allowed_tools=None, expires_at=None):
+        """Generate a new bearer token, store its hash, and return (instance, plaintext).
+
+        The plaintext is returned exactly once and is never persisted. Callers
+        must surface it to the human and rely on the user to copy it; after
+        this method returns, the plaintext is unrecoverable from the database.
+        """
+        plaintext = secrets.token_urlsafe(48)
+        instance = self.create(
+            user=user,
+            name=name,
+            token_hash=hash_token(plaintext),
+            allowed_tools=list(allowed_tools or []),
+            expires_at=expires_at,
+        )
+        return instance, plaintext
+
+
 class MCPToken(models.Model):
-    """Bearer token for authenticating MCP tool calls. See docs/Pattern_Django-MCP_V1-00.md."""
+    """Bearer token for authenticating MCP tool calls.
+
+    Tokens are hashed at rest (SHA-256, 64-char hex). Plaintext exists only in
+    memory at creation time, on the wire to the client, and in the user's own
+    storage. A leaked database backup discloses no usable credentials.
+    """
 
     user = models.ForeignKey(
         settings.AUTH_USER_MODEL,
         on_delete=models.CASCADE,
         related_name="mcp_tokens",
     )
-    token = models.CharField(max_length=64, unique=True, db_index=True)
+    token_hash = models.CharField(max_length=64, unique=True, db_index=True)
     name = models.CharField(max_length=100)
     is_active = models.BooleanField(default=True)
     expires_at = models.DateTimeField(null=True, blank=True)
@@ -22,17 +52,14 @@ class MCPToken(models.Model):
     created_at = models.DateTimeField(auto_now_add=True)
     updated_at = models.DateTimeField(auto_now=True)
 
+    objects = MCPTokenManager()
+
     class Meta:
         ordering = ["-created_at"]
 
     def __str__(self):
         return f"{self.name} ({self.user})"
 
-    def save(self, **kwargs):
-        if not self.token:
-            self.token = secrets.token_urlsafe(48)
-        super().save(**kwargs)
-
     @property
     def is_valid(self) -> bool:
         if not self.is_active:
@@ -51,6 +78,9 @@ class MCPToken(models.Model):
         self.save(update_fields=["last_used_at"])
 
     def get_masked_token(self) -> str:
-        if len(self.token) > 8:
-            return f"{'*' * (len(self.token) - 8)}{self.token[-8:]}"
-        return "*" * len(self.token)
+        """Token-id-style display for admin and dashboard.
+
+        Plaintext is unrecoverable, so we display the first 8 chars of the
+        hash prefixed with `mcp_…`. Stable per token, never reveals plaintext.
+        """
+        return f"mcp_…{self.token_hash[:8]}"

mnemosyne/mcp_server/server.py

@@ -5,7 +5,11 @@ from __future__ import annotations
 from fastmcp import FastMCP
 
 from .auth import MCPAuthMiddleware
-from .tools import register_discovery_tools, register_search_tools
+from .tools import (
+    register_discovery_tools,
+    register_health_tools,
+    register_search_tools,
+)
 
 INSTRUCTIONS = """\
 Mnemosyne is a content-type-aware, multimodal knowledge base. It indexes
@@ -23,6 +27,8 @@ shapes how content is chunked, embedded, and re-ranked:
 - film         — Scripts, synopses, stills.
 - art          — Catalogs, descriptions, artwork itself.
 - journal      — Personal entries; temporal/reflective.
+- business     — Proposals, marketing, sales, strategy. Commercial context.
+- finance      — Statements, tax, market commentary. Quote figures exactly.
 
 Tools:
 - search                      Hybrid retrieval. Filter by library_uid, library_type,
@@ -32,6 +38,7 @@ Tools:
 - list_libraries              Discover libraries (and their library_type).
 - list_collections            Discover collections, optionally per library.
 - list_items                  Discover indexed items (documents).
+- get_health                  Health check (used by Pallas/Daedalus pollers).
 
 Workflow: list_libraries → search(query, library_type=...) → get_chunk(chunk_uid)
 when the preview isn't enough. The calling LLM is responsible for synthesis
@@ -47,6 +54,7 @@ def build_server() -> FastMCP:
     mcp.add_middleware(MCPAuthMiddleware())
     register_search_tools(mcp)
     register_discovery_tools(mcp)
+    register_health_tools(mcp)
     return mcp
 
 

mnemosyne/mcp_server/templates/mcp_server/tokens/create.html

@@ -0,0 +1,63 @@
+{% extends "themis/base.html" %}
+
+{% block title %}Generate MCP Token — {{ themis_app_name }}{% endblock %}
+
+{% block content %}
+<div class="max-w-2xl mx-auto">
+    <h1 class="text-2xl font-bold mb-6">Generate MCP Token</h1>
+
+    <div class="alert alert-info mb-6">
+        <span>The token will be displayed once after creation. Save it before leaving the page — it cannot be recovered.</span>
+    </div>
+
+    <form method="post" action="{% url 'mcp_server:mcp-token-create' %}">
+        {% csrf_token %}
+
+        <div class="card bg-base-200 mb-6">
+            <div class="card-body">
+                <div class="form-control">
+                    <label class="label" for="id_name">
+                        <span class="label-text">Name</span>
+                    </label>
+                    {{ form.name }}
+                    <label class="label">
+                        <span class="label-text-alt opacity-60">A friendly label so you can identify this token later (e.g. “Claude Desktop”).</span>
+                    </label>
+                </div>
+                <div class="form-control mt-4">
+                    <label class="label" for="id_expires_at">
+                        <span class="label-text">Expires at (optional)</span>
+                    </label>
+                    {{ form.expires_at }}
+                    <label class="label">
+                        <span class="label-text-alt opacity-60">{{ form.expires_at.help_text }}</span>
+                    </label>
+                </div>
+            </div>
+        </div>
+
+        <div class="card bg-base-200 mb-6">
+            <div class="card-body">
+                <h2 class="card-title text-lg">Allowed Tools</h2>
+                <p class="text-sm opacity-60 mb-2">{{ form.allowed_tools.help_text }}</p>
+                <div class="space-y-1">
+                    {% for choice in form.allowed_tools %}
+                    <label class="label cursor-pointer justify-start gap-3">
+                        {{ choice.tag }}
+                        <span class="label-text font-mono">{{ choice.choice_label }}</span>
+                    </label>
+                    {% endfor %}
+                </div>
+                {% if form.allowed_tools.errors %}
+                <div class="text-error text-sm mt-2">{{ form.allowed_tools.errors }}</div>
+                {% endif %}
+            </div>
+        </div>
+
+        <div class="flex justify-between">
+            <a href="{% url 'mcp_server:mcp-token-list' %}" class="btn btn-ghost">Cancel</a>
+            <button type="submit" class="btn btn-primary">Generate Token</button>
+        </div>
+    </form>
+</div>
+{% endblock %}

mnemosyne/mcp_server/templates/mcp_server/tokens/created.html

@@ -0,0 +1,50 @@
+{% extends "themis/base.html" %}
+
+{% block title %}MCP Token Created — {{ themis_app_name }}{% endblock %}
+
+{% block content %}
+<div class="max-w-2xl mx-auto">
+    <h1 class="text-2xl font-bold mb-2">MCP Token Created</h1>
+    <p class="opacity-60 mb-6">{{ token.name }}</p>
+
+    <div class="alert alert-warning mb-6">
+        <span><strong>Save this token now.</strong> Once you leave this page it cannot be retrieved — only the hash is stored. If lost, you will need to generate a new one.</span>
+    </div>
+
+    <div class="card bg-base-200 mb-6">
+        <div class="card-body">
+            <h2 class="card-title text-lg mb-2">Token</h2>
+            <div class="font-mono break-all bg-base-300 p-4 rounded select-all" id="mcp-plaintext">{{ plaintext }}</div>
+            <div class="card-actions justify-end mt-3">
+                <button type="button" id="mcp-copy-btn" class="btn btn-sm btn-primary"
+                        onclick="(()=>{const t=document.getElementById('mcp-plaintext').textContent;navigator.clipboard.writeText(t).then(()=>{const b=document.getElementById('mcp-copy-btn');b.textContent='Copied!';setTimeout(()=>b.textContent='Copy to clipboard',2000);});})()">
+                    Copy to clipboard
+                </button>
+            </div>
+        </div>
+    </div>
+
+    <div class="card bg-base-200 mb-6">
+        <div class="card-body">
+            <h2 class="card-title text-lg">Use it</h2>
+            <p class="text-sm opacity-80 mb-3">
+                Add the token to your MCP client config as a Bearer credential. Example for Claude Desktop:
+            </p>
+<pre class="bg-base-300 p-4 rounded text-xs overflow-x-auto"><code>{
+  "mcpServers": {
+    "mnemosyne": {
+      "url": "http://localhost:8001/mcp/",
+      "headers": {
+        "Authorization": "Bearer &lt;paste token here&gt;"
+      }
+    }
+  }
+}</code></pre>
+        </div>
+    </div>
+
+    <div class="flex justify-end">
+        <a href="{% url 'mcp_server:mcp-token-list' %}" class="btn btn-primary">I’ve saved it — go to token list</a>
+    </div>
+</div>
+{% endblock %}

mnemosyne/mcp_server/templates/mcp_server/tokens/detail.html

@@ -0,0 +1,81 @@
+{% extends "themis/base.html" %}
+{% load humanize %}
+
+{% block title %}{{ token.name }} — {{ themis_app_name }}{% endblock %}
+
+{% block content %}
+<div class="max-w-2xl mx-auto">
+    <div class="flex items-center justify-between mb-6">
+        <h1 class="text-2xl font-bold">{{ token.name }}</h1>
+        <div class="flex gap-2">
+            <a href="{% url 'mcp_server:mcp-token-edit' pk=token.pk %}" class="btn btn-ghost btn-sm">Edit</a>
+            {% if token.is_active %}
+            <form method="post" action="{% url 'mcp_server:mcp-token-revoke' pk=token.pk %}"
+                  onsubmit="return confirm('Revoke this token? It will no longer authenticate MCP requests.');">
+                {% csrf_token %}
+                <button type="submit" class="btn btn-warning btn-sm btn-outline">Revoke</button>
+            </form>
+            {% endif %}
+            <form method="post" action="{% url 'mcp_server:mcp-token-delete' pk=token.pk %}"
+                  onsubmit="return confirm('Delete this token permanently? This removes the audit trail.');">
+                {% csrf_token %}
+                <button type="submit" class="btn btn-error btn-sm btn-outline">Delete</button>
+            </form>
+        </div>
+    </div>
+
+    <div class="card bg-base-200 mb-6">
+        <div class="card-body">
+            <div class="grid grid-cols-2 gap-y-3">
+                <div class="opacity-60">Token ID</div>
+                <div class="font-mono">{{ token.get_masked_token }}</div>
+
+                <div class="opacity-60">Status</div>
+                <div>
+                    {% if token.is_valid %}
+                    <span class="badge badge-success badge-sm">Active</span>
+                    {% elif not token.is_active %}
+                    <span class="badge badge-error badge-sm">Revoked</span>
+                    {% else %}
+                    <span class="badge badge-error badge-sm">Expired</span>
+                    {% endif %}
+                </div>
+
+                {% if token.expires_at %}
+                <div class="opacity-60">Expires</div>
+                <div>{{ token.expires_at }}</div>
+                {% endif %}
+
+                <div class="opacity-60">Last Used</div>
+                <div>
+                    {% if token.last_used_at %}
+                    {{ token.last_used_at }} <span class="opacity-60">({{ token.last_used_at|naturaltime }})</span>
+                    {% else %}
+                    <span class="opacity-60">Never</span>
+                    {% endif %}
+                </div>
+
+                <div class="opacity-60">Created</div>
+                <div>{{ token.created_at }}</div>
+            </div>
+        </div>
+    </div>
+
+    <div class="card bg-base-200 mb-6">
+        <div class="card-body">
+            <h2 class="card-title text-lg">Allowed Tools</h2>
+            {% if token.allowed_tools %}
+            <div class="flex flex-wrap gap-2">
+                {% for tool in token.allowed_tools %}
+                <span class="badge badge-outline font-mono">{{ tool }}</span>
+                {% endfor %}
+            </div>
+            {% else %}
+            <p class="opacity-60">All tools permitted.</p>
+            {% endif %}
+        </div>
+    </div>
+
+    <a href="{% url 'mcp_server:mcp-token-list' %}" class="btn btn-ghost btn-sm">← Back to Tokens</a>
+</div>
+{% endblock %}

mnemosyne/mcp_server/templates/mcp_server/tokens/edit.html

@@ -0,0 +1,60 @@
+{% extends "themis/base.html" %}
+
+{% block title %}Edit {{ token.name }} — {{ themis_app_name }}{% endblock %}
+
+{% block content %}
+<div class="max-w-2xl mx-auto">
+    <h1 class="text-2xl font-bold mb-6">Edit Token: {{ token.name }}</h1>
+
+    <div class="alert alert-info mb-6">
+        <span>You can edit metadata below. The token value itself cannot be changed — generate a new token if needed.</span>
+    </div>
+
+    <form method="post" action="{% url 'mcp_server:mcp-token-edit' pk=token.pk %}">
+        {% csrf_token %}
+
+        <div class="card bg-base-200 mb-6">
+            <div class="card-body">
+                <div class="form-control">
+                    <label class="label" for="id_name">
+                        <span class="label-text">Name</span>
+                    </label>
+                    {{ form.name }}
+                </div>
+                <div class="form-control mt-4">
+                    <label class="label cursor-pointer justify-start gap-3">
+                        {{ form.is_active }}
+                        <span class="label-text">Active</span>
+                    </label>
+                </div>
+                <div class="form-control mt-4">
+                    <label class="label" for="id_expires_at">
+                        <span class="label-text">Expires at (optional)</span>
+                    </label>
+                    {{ form.expires_at }}
+                </div>
+            </div>
+        </div>
+
+        <div class="card bg-base-200 mb-6">
+            <div class="card-body">
+                <h2 class="card-title text-lg">Allowed Tools</h2>
+                <p class="text-sm opacity-60 mb-2">{{ form.allowed_tools.help_text }}</p>
+                <div class="space-y-1">
+                    {% for choice in form.allowed_tools %}
+                    <label class="label cursor-pointer justify-start gap-3">
+                        {{ choice.tag }}
+                        <span class="label-text font-mono">{{ choice.choice_label }}</span>
+                    </label>
+                    {% endfor %}
+                </div>
+            </div>
+        </div>
+
+        <div class="flex justify-between">
+            <a href="{% url 'mcp_server:mcp-token-detail' pk=token.pk %}" class="btn btn-ghost">Cancel</a>
+            <button type="submit" class="btn btn-primary">Save Changes</button>
+        </div>
+    </form>
+</div>
+{% endblock %}

mnemosyne/mcp_server/templates/mcp_server/tokens/list.html

@@ -0,0 +1,61 @@
+{% extends "themis/base.html" %}
+{% load humanize %}
+
+{% block title %}MCP Tokens — {{ themis_app_name }}{% endblock %}
+
+{% block content %}
+<div class="max-w-3xl mx-auto">
+    <div class="flex items-center justify-between mb-6">
+        <h1 class="text-2xl font-bold">MCP Tokens</h1>
+        <a href="{% url 'mcp_server:mcp-token-create' %}" class="btn btn-primary btn-sm">Generate Token</a>
+    </div>
+
+    <p class="text-sm opacity-60 mb-4">
+        Bearer tokens used by MCP clients (Claude Desktop, Cursor, etc.) to call Mnemosyne. Stored hashed at rest — the plaintext is shown only once at creation.
+    </p>
+
+    {% if tokens %}
+    <div class="space-y-3">
+        {% for token in tokens %}
+        <div class="card bg-base-200">
+            <div class="card-body p-4">
+                <div class="flex items-center justify-between">
+                    <div>
+                        <a href="{% url 'mcp_server:mcp-token-detail' pk=token.pk %}"
+                           class="font-semibold link link-hover">
+                            {{ token.name }}
+                        </a>
+                        <div class="flex items-center gap-2 mt-1 flex-wrap">
+                            <span class="text-sm opacity-60 font-mono">{{ token.get_masked_token }}</span>
+                            {% if not token.is_valid %}
+                            <span class="badge badge-sm badge-error">
+                                {% if not token.is_active %}Revoked{% else %}Expired{% endif %}
+                            </span>
+                            {% else %}
+                            <span class="badge badge-sm badge-success">Active</span>
+                            {% endif %}
+                            <span class="text-xs opacity-60">
+                                Last used:
+                                {% if token.last_used_at %}{{ token.last_used_at|naturaltime }}{% else %}never{% endif %}
+                            </span>
+                        </div>
+                    </div>
+                    <div class="flex gap-1">
+                        <a href="{% url 'mcp_server:mcp-token-edit' pk=token.pk %}"
+                           class="btn btn-ghost btn-xs">Edit</a>
+                    </div>
+                </div>
+            </div>
+        </div>
+        {% endfor %}
+    </div>
+    {% else %}
+    <div class="card bg-base-200">
+        <div class="card-body items-center text-center py-12">
+            <p class="opacity-60 mb-4">No MCP tokens yet.</p>
+            <a href="{% url 'mcp_server:mcp-token-create' %}" class="btn btn-primary btn-sm">Generate Your First Token</a>
+        </div>
+    </div>
+    {% endif %}
+</div>
+{% endblock %}

mnemosyne/mcp_server/tests/test_auth.py

@@ -17,16 +17,18 @@ class ResolveMCPUserTest(TestCase):
         self.user = User.objects.create_user(
             username="bob", email="bob@example.com", password="pw"
         )
-        self.token = MCPToken.objects.create(user=self.user, name="t")
+        self.token, self.plaintext = MCPToken.objects.create_token(
+            user=self.user, name="t"
+        )
 
     def test_resolves_valid_token(self):
-        user, token = resolve_mcp_user(self.token.token)
+        user, token = resolve_mcp_user(self.plaintext)
         self.assertEqual(user.pk, self.user.pk)
         self.assertEqual(token.pk, self.token.pk)
 
     def test_records_usage(self):
         self.assertIsNone(self.token.last_used_at)
-        resolve_mcp_user(self.token.token)
+        resolve_mcp_user(self.plaintext)
         self.token.refresh_from_db()
         self.assertIsNotNone(self.token.last_used_at)
 
@@ -38,16 +40,31 @@ class ResolveMCPUserTest(TestCase):
         self.token.is_active = False
         self.token.save()
         with self.assertRaises(MCPAuthError):
-            resolve_mcp_user(self.token.token)
+            resolve_mcp_user(self.plaintext)
 
     def test_expired_token_raises(self):
         self.token.expires_at = timezone.now() - timedelta(hours=1)
         self.token.save()
         with self.assertRaises(MCPAuthError):
-            resolve_mcp_user(self.token.token)
+            resolve_mcp_user(self.plaintext)
 
     def test_disabled_user_raises(self):
         self.user.is_active = False
         self.user.save()
         with self.assertRaises(MCPAuthError):
-            resolve_mcp_user(self.token.token)
+            resolve_mcp_user(self.plaintext)
+
+    def test_plaintext_not_in_db(self):
+        # Defense in depth: scan every column for the plaintext value.
+        from django.db import connection
+
+        plaintext = self.plaintext
+        with connection.cursor() as cur:
+            cur.execute("SELECT * FROM mcp_server_mcptoken")
+            rows = cur.fetchall()
+        for row in rows:
+            for value in row:
+                self.assertNotEqual(
+                    value, plaintext,
+                    f"Plaintext token leaked into the database: {value!r}",
+                )

mnemosyne/mcp_server/tests/test_forms.py

@@ -0,0 +1,58 @@
+"""Form tests for the MCP token dashboard."""
+
+from django.contrib.auth import get_user_model
+from django.test import TestCase
+
+from mcp_server.forms import MCPTokenCreateForm, MCPTokenEditForm
+from mcp_server.models import MCPToken
+
+User = get_user_model()
+
+
+class CreateFormTest(TestCase):
+    def test_required_fields(self):
+        form = MCPTokenCreateForm(data={})
+        self.assertFalse(form.is_valid())
+        self.assertIn("name", form.errors)
+
+    def test_name_only_is_valid(self):
+        form = MCPTokenCreateForm(data={"name": "Test"})
+        self.assertTrue(form.is_valid(), form.errors)
+
+    def test_tool_choices_match_registered_tools(self):
+        form = MCPTokenCreateForm()
+        choices = {value for value, _ in form.fields["allowed_tools"].choices}
+        # These five must always be present per the FastMCP server.
+        for expected in {"search", "get_chunk", "list_libraries", "list_collections", "list_items"}:
+            self.assertIn(expected, choices)
+
+
+class EditFormTest(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username="alice", password="pw")
+        self.token, _ = MCPToken.objects.create_token(
+            user=self.user, name="t", allowed_tools=["search"]
+        )
+
+    def test_initial_allowed_tools_populated(self):
+        form = MCPTokenEditForm(instance=self.token)
+        self.assertEqual(form.fields["allowed_tools"].initial, ["search"])
+
+    def test_save_updates_metadata(self):
+        form = MCPTokenEditForm(
+            data={
+                "name": "Renamed",
+                "is_active": False,
+                "expires_at": "",
+                "allowed_tools": ["search", "get_chunk"],
+            },
+            instance=self.token,
+        )
+        self.assertTrue(form.is_valid(), form.errors)
+        instance = form.save(commit=False)
+        instance.allowed_tools = form.cleaned_data["allowed_tools"]
+        instance.save()
+        self.token.refresh_from_db()
+        self.assertEqual(self.token.name, "Renamed")
+        self.assertFalse(self.token.is_active)
+        self.assertEqual(self.token.allowed_tools, ["search", "get_chunk"])

mnemosyne/mcp_server/tests/test_server.py

@@ -7,7 +7,14 @@ from django.test import TestCase
 from mcp_server.server import mcp
 
 
-EXPECTED_TOOLS = {"search", "get_chunk", "list_libraries", "list_collections", "list_items"}
+EXPECTED_TOOLS = {
+    "search",
+    "get_chunk",
+    "list_libraries",
+    "list_collections",
+    "list_items",
+    "get_health",
+}
 
 
 class ServerRegistrationTest(TestCase):

mnemosyne/mcp_server/tests/test_token.py

@@ -6,7 +6,7 @@ from django.contrib.auth import get_user_model
 from django.test import TestCase
 from django.utils import timezone
 
-from mcp_server.models import MCPToken
+from mcp_server.models import MCPToken, hash_token
 
 User = get_user_model()
 
@@ -17,21 +17,33 @@ class MCPTokenModelTest(TestCase):
             username="alice", email="alice@example.com", password="pw"
         )
 
-    def test_token_auto_generated(self):
-        token = MCPToken.objects.create(user=self.user, name="t")
-        self.assertTrue(token.token)
-        self.assertGreater(len(token.token), 20)
+    def test_create_token_returns_plaintext_and_stores_hash(self):
+        token, plaintext = MCPToken.objects.create_token(user=self.user, name="t")
+        self.assertTrue(plaintext)
+        self.assertGreater(len(plaintext), 20)
+        # Database stores hash, not plaintext
+        self.assertEqual(len(token.token_hash), 64)
+        self.assertNotEqual(token.token_hash, plaintext)
+        self.assertEqual(token.token_hash, hash_token(plaintext))
+
+    def test_token_hash_never_equals_plaintext(self):
+        # Regression guard: if anyone ever wires plaintext back into token_hash,
+        # this fails.
+        token, plaintext = MCPToken.objects.create_token(user=self.user, name="t")
+        self.assertNotIn(plaintext, token.token_hash)
 
     def test_active_token_is_valid(self):
-        token = MCPToken.objects.create(user=self.user, name="t")
+        token, _ = MCPToken.objects.create_token(user=self.user, name="t")
         self.assertTrue(token.is_valid)
 
     def test_inactive_token_not_valid(self):
-        token = MCPToken.objects.create(user=self.user, name="t", is_active=False)
+        token, _ = MCPToken.objects.create_token(user=self.user, name="t")
+        token.is_active = False
+        token.save()
         self.assertFalse(token.is_valid)
 
     def test_expired_token_not_valid(self):
-        token = MCPToken.objects.create(
+        token, _ = MCPToken.objects.create_token(
             user=self.user,
             name="t",
             expires_at=timezone.now() - timedelta(hours=1),
@@ -39,25 +51,27 @@ class MCPTokenModelTest(TestCase):
         self.assertFalse(token.is_valid)
 
     def test_unrestricted_permits_all(self):
-        token = MCPToken.objects.create(user=self.user, name="t")
+        token, _ = MCPToken.objects.create_token(user=self.user, name="t")
         self.assertTrue(token.can_use_tool("anything"))
 
     def test_tool_whitelist(self):
-        token = MCPToken.objects.create(
+        token, _ = MCPToken.objects.create_token(
             user=self.user, name="t", allowed_tools=["search"]
         )
         self.assertTrue(token.can_use_tool("search"))
         self.assertFalse(token.can_use_tool("get_chunk"))
 
     def test_record_usage(self):
-        token = MCPToken.objects.create(user=self.user, name="t")
+        token, _ = MCPToken.objects.create_token(user=self.user, name="t")
         self.assertIsNone(token.last_used_at)
         token.record_usage()
         token.refresh_from_db()
         self.assertIsNotNone(token.last_used_at)
 
-    def test_masked_token(self):
-        token = MCPToken.objects.create(user=self.user, name="t")
+    def test_masked_token_is_hash_prefix(self):
+        token, plaintext = MCPToken.objects.create_token(user=self.user, name="t")
         masked = token.get_masked_token()
-        self.assertTrue(masked.endswith(token.token[-8:]))
-        self.assertIn("*", masked)
+        self.assertTrue(masked.startswith("mcp_…"))
+        self.assertIn(token.token_hash[:8], masked)
+        # Plaintext must never leak through the masked display
+        self.assertNotIn(plaintext, masked)

mnemosyne/mcp_server/tests/test_views.py

@@ -0,0 +1,177 @@
+"""View tests for the MCP token self-service dashboard."""
+
+from django.contrib.auth import get_user_model
+from django.test import TestCase
+from django.urls import reverse
+
+from mcp_server.models import MCPToken
+
+User = get_user_model()
+
+
+class TokenListViewTest(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(
+            username="alice", email="alice@example.com", password="pw"
+        )
+        self.url = reverse("mcp_server:mcp-token-list")
+
+    def test_login_required(self):
+        resp = self.client.get(self.url)
+        self.assertEqual(resp.status_code, 302)
+        self.assertIn("/login/", resp.url)
+
+    def test_list_shows_only_own_tokens(self):
+        other = User.objects.create_user(username="bob", password="pw")
+        MCPToken.objects.create_token(user=self.user, name="mine")
+        MCPToken.objects.create_token(user=other, name="theirs")
+        self.client.force_login(self.user)
+        resp = self.client.get(self.url)
+        self.assertContains(resp, "mine")
+        self.assertNotContains(resp, "theirs")
+
+    def test_empty_state(self):
+        self.client.force_login(self.user)
+        resp = self.client.get(self.url)
+        self.assertContains(resp, "No MCP tokens yet.")
+
+
+class TokenCreateViewTest(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username="alice", password="pw")
+        self.client.force_login(self.user)
+        self.url = reverse("mcp_server:mcp-token-create")
+
+    def test_get_renders_form(self):
+        resp = self.client.get(self.url)
+        self.assertEqual(resp.status_code, 200)
+        self.assertContains(resp, "Generate MCP Token")
+
+    def test_post_creates_token_and_shows_plaintext_once(self):
+        resp = self.client.post(self.url, {"name": "Claude Desktop"})
+        self.assertEqual(resp.status_code, 200)
+        self.assertContains(resp, "Save this token now")
+        # Pull the created row, verify the response contained a plaintext that
+        # is NOT what we stored.
+        token = MCPToken.objects.get(user=self.user, name="Claude Desktop")
+        self.assertNotContains(resp, token.token_hash)  # hash is not what we display
+        # And the detail page never renders the plaintext.
+        body = resp.content.decode()
+        # Find the plaintext from the response: the only long alphanumeric
+        # block inside the #mcp-plaintext div.
+        import re
+
+        m = re.search(r'id="mcp-plaintext">([A-Za-z0-9_\-]+)<', body)
+        self.assertIsNotNone(m, "plaintext block not found in response")
+        plaintext = m.group(1)
+        # Sanity: round-tripping the plaintext through hash_token reproduces
+        # what's stored.
+        from mcp_server.models import hash_token
+
+        self.assertEqual(hash_token(plaintext), token.token_hash)
+        # Detail page must NOT contain the plaintext.
+        detail_resp = self.client.get(
+            reverse("mcp_server:mcp-token-detail", args=[token.pk])
+        )
+        self.assertNotContains(detail_resp, plaintext)
+
+    def test_post_invalid_renders_form_again(self):
+        resp = self.client.post(self.url, {"name": ""})
+        self.assertEqual(resp.status_code, 200)
+        self.assertContains(resp, "Generate MCP Token")
+        self.assertEqual(MCPToken.objects.count(), 0)
+
+
+class TokenDetailViewTest(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username="alice", password="pw")
+        self.client.force_login(self.user)
+        self.token, _ = MCPToken.objects.create_token(user=self.user, name="t")
+
+    def test_renders_token(self):
+        resp = self.client.get(
+            reverse("mcp_server:mcp-token-detail", args=[self.token.pk])
+        )
+        self.assertContains(resp, self.token.name)
+        self.assertContains(resp, self.token.get_masked_token())
+
+    def test_cannot_view_other_users_token(self):
+        other = User.objects.create_user(username="bob", password="pw")
+        other_token, _ = MCPToken.objects.create_token(user=other, name="theirs")
+        resp = self.client.get(
+            reverse("mcp_server:mcp-token-detail", args=[other_token.pk])
+        )
+        self.assertEqual(resp.status_code, 404)
+
+
+class TokenEditViewTest(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username="alice", password="pw")
+        self.client.force_login(self.user)
+        self.token, _ = MCPToken.objects.create_token(user=self.user, name="t")
+
+    def test_post_updates_metadata(self):
+        resp = self.client.post(
+            reverse("mcp_server:mcp-token-edit", args=[self.token.pk]),
+            {
+                "name": "Renamed",
+                "is_active": "on",
+                "expires_at": "",
+                "allowed_tools": ["search"],
+            },
+        )
+        self.assertEqual(resp.status_code, 302)
+        self.token.refresh_from_db()
+        self.assertEqual(self.token.name, "Renamed")
+        self.assertEqual(self.token.allowed_tools, ["search"])
+
+    def test_cannot_edit_other_users_token(self):
+        other = User.objects.create_user(username="bob", password="pw")
+        other_token, _ = MCPToken.objects.create_token(user=other, name="theirs")
+        resp = self.client.post(
+            reverse("mcp_server:mcp-token-edit", args=[other_token.pk]),
+            {"name": "hacked"},
+        )
+        self.assertEqual(resp.status_code, 404)
+
+
+class TokenRevokeViewTest(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username="alice", password="pw")
+        self.client.force_login(self.user)
+        self.token, _ = MCPToken.objects.create_token(user=self.user, name="t")
+
+    def test_revoke_sets_inactive_keeps_row(self):
+        url = reverse("mcp_server:mcp-token-revoke", args=[self.token.pk])
+        resp = self.client.post(url)
+        self.assertEqual(resp.status_code, 302)
+        self.token.refresh_from_db()
+        self.assertFalse(self.token.is_active)
+        # Row still exists for audit trail.
+        self.assertTrue(MCPToken.objects.filter(pk=self.token.pk).exists())
+
+    def test_get_not_allowed(self):
+        url = reverse("mcp_server:mcp-token-revoke", args=[self.token.pk])
+        resp = self.client.get(url)
+        self.assertEqual(resp.status_code, 405)
+
+
+class TokenDeleteViewTest(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username="alice", password="pw")
+        self.client.force_login(self.user)
+        self.token, _ = MCPToken.objects.create_token(user=self.user, name="t")
+
+    def test_delete_removes_row(self):
+        url = reverse("mcp_server:mcp-token-delete", args=[self.token.pk])
+        resp = self.client.post(url)
+        self.assertEqual(resp.status_code, 302)
+        self.assertFalse(MCPToken.objects.filter(pk=self.token.pk).exists())
+
+    def test_cannot_delete_other_users_token(self):
+        other = User.objects.create_user(username="bob", password="pw")
+        other_token, _ = MCPToken.objects.create_token(user=other, name="theirs")
+        url = reverse("mcp_server:mcp-token-delete", args=[other_token.pk])
+        resp = self.client.post(url)
+        self.assertEqual(resp.status_code, 404)
+        self.assertTrue(MCPToken.objects.filter(pk=other_token.pk).exists())

mnemosyne/mcp_server/tools/__init__.py

@@ -1,4 +1,9 @@
 from .discovery import register_discovery_tools
+from .health import register_health_tools
 from .search import register_search_tools
 
-__all__ = ["register_search_tools", "register_discovery_tools"]
+__all__ = [
+    "register_search_tools",
+    "register_discovery_tools",
+    "register_health_tools",
+]

mnemosyne/mcp_server/tools/discovery.py

@@ -20,16 +20,21 @@ def _clamp(limit: int) -> int:
 
 def register_discovery_tools(mcp):
     @mcp.tool
-    async def list_libraries(limit: int = DEFAULT_LIMIT, offset: int = 0) -> dict[str, Any]:
+    async def list_libraries(
+        limit: int = DEFAULT_LIMIT,
+        offset: int = 0,
+        # System-injected; deliberately absent from the docstring.
+        workspace_id: str | None = None,
+    ) -> dict[str, Any]:
         """List Mnemosyne libraries. Each library has a content-aware library_type
-        (fiction, nonfiction, technical, music, film, art, journal) that drives
-        chunking, embedding, and re-ranking. Returns uid, name, library_type,
-        description for each library — use the uid or library_type to scope a
-        subsequent search.
+        (fiction, nonfiction, technical, music, film, art, journal, business,
+        finance) that drives chunking, embedding, and re-ranking. Returns uid,
+        name, library_type, description for each library — use the uid or
+        library_type to scope a subsequent search.
         """
         with record_tool_call("list_libraries"):
             return await sync_to_async(_query_libraries, thread_sensitive=True)(
-                _clamp(limit), max(offset, 0)
+                _clamp(limit), max(offset, 0), workspace_id
             )
 
     @mcp.tool
@@ -37,6 +42,8 @@ def register_discovery_tools(mcp):
         library_uid: str | None = None,
         limit: int = DEFAULT_LIMIT,
         offset: int = 0,
+        # System-injected; deliberately absent from the docstring.
+        workspace_id: str | None = None,
     ) -> dict[str, Any]:
         """List collections, optionally filtered by parent library_uid.
         Collections group related items inside a library (e.g. a series of novels,
@@ -45,7 +52,7 @@ def register_discovery_tools(mcp):
         """
         with record_tool_call("list_collections"):
             return await sync_to_async(_query_collections, thread_sensitive=True)(
-                library_uid, _clamp(limit), max(offset, 0)
+                library_uid, _clamp(limit), max(offset, 0), workspace_id
             )
 
     @mcp.tool
@@ -54,6 +61,8 @@ def register_discovery_tools(mcp):
         library_uid: str | None = None,
         limit: int = DEFAULT_LIMIT,
         offset: int = 0,
+        # System-injected; deliberately absent from the docstring.
+        workspace_id: str | None = None,
     ) -> dict[str, Any]:
         """List items (the indexed documents/files), optionally filtered by
         collection_uid or library_uid. Returns uid, title, item_type, file_type,
@@ -63,17 +72,27 @@ def register_discovery_tools(mcp):
         """
         with record_tool_call("list_items"):
             return await sync_to_async(_query_items, thread_sensitive=True)(
-                collection_uid, library_uid, _clamp(limit), max(offset, 0)
+                collection_uid, library_uid, _clamp(limit), max(offset, 0), workspace_id
             )
 
 
-def _query_libraries(limit: int, offset: int) -> dict[str, Any]:
+_WORKSPACE_SCOPE = (
+    "($workspace_id IS NULL AND l.workspace_id IS NULL OR "
+    "l.workspace_id = $workspace_id)"
+)
+
+
+def _query_libraries(
+    limit: int, offset: int, workspace_id: str | None = None
+) -> dict[str, Any]:
     from neomodel import db
 
     rows, _ = db.cypher_query(
-        "MATCH (l:Library) RETURN l.uid, l.name, l.library_type, l.description "
+        "MATCH (l:Library) "
+        f"WHERE {_WORKSPACE_SCOPE} "
+        "RETURN l.uid, l.name, l.library_type, l.description "
         "ORDER BY l.name SKIP $offset LIMIT $limit",
-        {"offset": offset, "limit": limit},
+        {"offset": offset, "limit": limit, "workspace_id": workspace_id},
     )
     return {
         "libraries": [
@@ -91,24 +110,33 @@ def _query_libraries(limit: int, offset: int) -> dict[str, Any]:
 
 
 def _query_collections(
-    library_uid: str | None, limit: int, offset: int
+    library_uid: str | None, limit: int, offset: int,
+    workspace_id: str | None = None,
 ) -> dict[str, Any]:
     from neomodel import db
 
     if library_uid:
         cypher = (
             "MATCH (l:Library {uid: $library_uid})-[:CONTAINS]->(c:Collection) "
+            f"WHERE {_WORKSPACE_SCOPE} "
             "RETURN c.uid, c.name, c.description, l.uid, l.name "
             "ORDER BY c.name SKIP $offset LIMIT $limit"
         )
-        params = {"library_uid": library_uid, "offset": offset, "limit": limit}
+        params = {
+            "library_uid": library_uid, "offset": offset, "limit": limit,
+            "workspace_id": workspace_id,
+        }
     else:
         cypher = (
             "MATCH (l:Library)-[:CONTAINS]->(c:Collection) "
+            f"WHERE {_WORKSPACE_SCOPE} "
             "RETURN c.uid, c.name, c.description, l.uid, l.name "
             "ORDER BY l.name, c.name SKIP $offset LIMIT $limit"
         )
-        params = {"offset": offset, "limit": limit}
+        params = {
+            "offset": offset, "limit": limit,
+            "workspace_id": workspace_id,
+        }
 
     rows, _ = db.cypher_query(cypher, params)
     return {
@@ -132,11 +160,15 @@ def _query_items(
     library_uid: str | None,
     limit: int,
     offset: int,
+    workspace_id: str | None = None,
 ) -> dict[str, Any]:
     from neomodel import db
 
-    where = []
-    params: dict[str, Any] = {"offset": offset, "limit": limit}
+    where = [_WORKSPACE_SCOPE]
+    params: dict[str, Any] = {
+        "offset": offset, "limit": limit,
+        "workspace_id": workspace_id,
+    }
     if collection_uid:
         where.append("c.uid = $collection_uid")
         params["collection_uid"] = collection_uid
@@ -144,7 +176,7 @@ def _query_items(
         where.append("l.uid = $library_uid")
         params["library_uid"] = library_uid
 
-    where_clause = ("WHERE " + " AND ".join(where)) if where else ""
+    where_clause = "WHERE " + " AND ".join(where)
     cypher = (
         "MATCH (l:Library)-[:CONTAINS]->(c:Collection)-[:CONTAINS]->(i:Item) "
         f"{where_clause} "

mnemosyne/mcp_server/tools/health.py

@@ -0,0 +1,123 @@
+"""Health-check MCP tool — used by Pallas/Daedalus health pollers.
+
+Per the Pallas health spec, returns one of:
+- ok        — all dependencies reachable
+- degraded  — non-critical dependency unhealthy (chat allowed)
+- error     — critical dependency unhealthy (chat blocked)
+
+The tool is intercepted by the FastMCP server and never invokes an LLM —
+it executes synchronously against Neo4j, S3, and the embedding model
+endpoint, and returns within the poller's timeout.
+"""
+
+from __future__ import annotations
+
+import time
+from typing import Any
+
+from asgiref.sync import sync_to_async
+
+from ..metrics import record_tool_call
+
+
+def register_health_tools(mcp):
+    @mcp.tool
+    async def get_health() -> dict[str, Any]:
+        """Health check for Mnemosyne.
+
+        Returns a status object compatible with the Pallas health spec:
+        {status: "ok"|"degraded"|"error", checks: {neo4j, s3, embedding}}.
+        """
+        with record_tool_call("get_health"):
+            return await sync_to_async(_run_health_check, thread_sensitive=True)()
+
+
+def _run_health_check() -> dict[str, Any]:
+    """Synchronous health check across Neo4j, S3, and embedding model."""
+    checks: dict[str, dict[str, Any]] = {}
+
+    checks["neo4j"] = _check_neo4j()
+    checks["s3"] = _check_s3()
+    checks["embedding"] = _check_embedding_model()
+
+    # Aggregate status: error if any critical check failed; degraded if a
+    # non-critical check failed; ok otherwise.
+    if checks["neo4j"]["status"] == "error" or checks["s3"]["status"] == "error":
+        status = "error"
+    elif any(c["status"] != "ok" for c in checks.values()):
+        status = "degraded"
+    else:
+        status = "ok"
+
+    return {
+        "status": status,
+        "checks": checks,
+    }
+
+
+def _check_neo4j() -> dict[str, Any]:
+    start = time.time()
+    try:
+        from neomodel import db
+
+        db.cypher_query("RETURN 1")
+        return {
+            "status": "ok",
+            "duration_ms": round((time.time() - start) * 1000, 1),
+        }
+    except Exception as exc:
+        return {
+            "status": "error",
+            "error": str(exc),
+            "duration_ms": round((time.time() - start) * 1000, 1),
+        }
+
+
+def _check_s3() -> dict[str, Any]:
+    start = time.time()
+    try:
+        from django.core.files.storage import default_storage
+
+        # `exists` on a path that won't exist is the cheapest round-trip
+        # we have. It returns False rather than raising on most backends.
+        default_storage.exists("__healthcheck__")
+        return {
+            "status": "ok",
+            "duration_ms": round((time.time() - start) * 1000, 1),
+        }
+    except Exception as exc:
+        return {
+            "status": "error",
+            "error": str(exc),
+            "duration_ms": round((time.time() - start) * 1000, 1),
+        }
+
+
+def _check_embedding_model() -> dict[str, Any]:
+    """Soft check: confirm a system embedding model is configured.
+
+    We don't hit the model endpoint here — that would burn GPU time on
+    every poll. The poller-level check is "is a model registered."
+    """
+    start = time.time()
+    try:
+        from llm_manager.models import LLMModel
+
+        model = LLMModel.get_system_embedding_model()
+        if model is None:
+            return {
+                "status": "degraded",
+                "error": "no system embedding model configured",
+                "duration_ms": round((time.time() - start) * 1000, 1),
+            }
+        return {
+            "status": "ok",
+            "model": model.name,
+            "duration_ms": round((time.time() - start) * 1000, 1),
+        }
+    except Exception as exc:
+        return {
+            "status": "degraded",
+            "error": str(exc),
+            "duration_ms": round((time.time() - start) * 1000, 1),
+        }

mnemosyne/mcp_server/tools/search.py

@@ -27,13 +27,19 @@ def register_search_tools(mcp):
         rerank: bool = True,
         include_images: bool = True,
         search_types: list[str] | None = None,
+        # workspace_id is system-injected by Daedalus's chat path. It is
+        # intentionally absent from the docstring so the calling LLM is
+        # never told it exists. Whatever value the LLM produces here is
+        # overwritten by Daedalus before the call reaches Mnemosyne.
+        workspace_id: str | None = None,
         ctx: Context | None = None,
     ) -> dict[str, Any]:
         """Hybrid retrieval over Mnemosyne: vector + full-text + concept-graph
         candidates fused by RRF and optionally re-ranked by Synesis.
 
         Filters: library_uid (exact library), library_type (one of fiction,
-        nonfiction, technical, music, film, art, journal), or collection_uid.
+        nonfiction, technical, music, film, art, journal, business, finance),
+        or collection_uid.
         Set rerank=False to skip re-ranking. search_types defaults to all three.
 
         Returns ranked candidates with chunk_uid (use get_chunk for full text),
@@ -49,6 +55,7 @@ def register_search_tools(mcp):
                 library_uid=library_uid,
                 library_type=library_type,
                 collection_uid=collection_uid,
+                workspace_id=workspace_id,
                 limit=limit,
                 rerank=rerank,
                 include_images=include_images,
@@ -56,7 +63,12 @@ def register_search_tools(mcp):
             )
 
     @mcp.tool
-    async def get_chunk(chunk_uid: str, ctx: Context | None = None) -> dict[str, Any]:
+    async def get_chunk(
+        chunk_uid: str,
+        # System-injected; deliberately absent from the docstring.
+        workspace_id: str | None = None,
+        ctx: Context | None = None,
+    ) -> dict[str, Any]:
         """Fetch the full text of a chunk by its uid (typically obtained from `search`).
 
         Returns the chunk text plus parent item context: chunk_uid, chunk_index,
@@ -64,11 +76,13 @@ def register_search_tools(mcp):
         text_preview from `search` isn't enough.
         """
         with record_tool_call("get_chunk"):
-            return await sync_to_async(_load_chunk, thread_sensitive=True)(chunk_uid)
+            return await sync_to_async(_load_chunk, thread_sensitive=True)(
+                chunk_uid, workspace_id
+            )
 
 
-def _run_search(*, user, query, library_uid, library_type, collection_uid, limit,
-                rerank, include_images, search_types) -> dict[str, Any]:
+def _run_search(*, user, query, library_uid, library_type, collection_uid,
+                workspace_id, limit, rerank, include_images, search_types) -> dict[str, Any]:
     from library.services.search import SearchRequest, SearchService
 
     req = SearchRequest(
@@ -76,6 +90,7 @@ def _run_search(*, user, query, library_uid, library_type, collection_uid, limit
         library_uid=library_uid,
         library_type=library_type,
         collection_uid=collection_uid,
+        workspace_id=workspace_id,
         search_types=search_types,
         limit=limit,
         vector_top_k=getattr(settings, "SEARCH_VECTOR_TOP_K", 50),
@@ -97,15 +112,17 @@ def _run_search(*, user, query, library_uid, library_type, collection_uid, limit
     }
 
 
-def _load_chunk(chunk_uid: str) -> dict[str, Any]:
+def _load_chunk(chunk_uid: str, workspace_id: str | None = None) -> dict[str, Any]:
     from neomodel import db
 
     rows, _ = db.cypher_query(
         "MATCH (l:Library)-[:CONTAINS]->(:Collection)-[:CONTAINS]->"
         "(i:Item)-[:HAS_CHUNK]->(c:Chunk {uid: $uid}) "
+        "WHERE ($workspace_id IS NULL AND l.workspace_id IS NULL OR "
+        "       l.workspace_id = $workspace_id) "
         "RETURN c.uid, c.chunk_index, c.chunk_s3_key, "
         "i.uid, i.title, l.library_type LIMIT 1",
-        {"uid": chunk_uid},
+        {"uid": chunk_uid, "workspace_id": workspace_id},
     )
     if not rows:
         raise ValueError(f"Chunk not found: {chunk_uid}")

mnemosyne/mcp_server/urls.py

@@ -0,0 +1,16 @@
+"""URL routes for the MCP token self-service dashboard."""
+
+from django.urls import path
+
+from . import views
+
+app_name = "mcp_server"
+
+urlpatterns = [
+    path("profile/mcp-tokens/", views.mcp_token_list, name="mcp-token-list"),
+    path("profile/mcp-tokens/add/", views.mcp_token_create, name="mcp-token-create"),
+    path("profile/mcp-tokens/<int:pk>/", views.mcp_token_detail, name="mcp-token-detail"),
+    path("profile/mcp-tokens/<int:pk>/edit/", views.mcp_token_edit, name="mcp-token-edit"),
+    path("profile/mcp-tokens/<int:pk>/revoke/", views.mcp_token_revoke, name="mcp-token-revoke"),
+    path("profile/mcp-tokens/<int:pk>/delete/", views.mcp_token_delete, name="mcp-token-delete"),
+]

mnemosyne/mcp_server/views.py

@@ -0,0 +1,95 @@
+"""Self-service dashboard for MCP bearer tokens.
+
+Mirrors the Themis API-keys flow visually but stores hashed tokens. Plaintext
+is shown to the user exactly once (on the create-success page) and never
+persisted.
+"""
+
+from __future__ import annotations
+
+from django.contrib import messages
+from django.contrib.auth.decorators import login_required
+from django.http import HttpRequest, HttpResponse
+from django.shortcuts import get_object_or_404, redirect, render
+from django.views.decorators.http import require_GET, require_http_methods, require_POST
+
+from .forms import MCPTokenCreateForm, MCPTokenEditForm
+from .models import MCPToken
+
+
+@login_required
+@require_GET
+def mcp_token_list(request: HttpRequest) -> HttpResponse:
+    tokens = MCPToken.objects.filter(user=request.user).order_by("-created_at")
+    return render(request, "mcp_server/tokens/list.html", {"tokens": tokens})
+
+
+@login_required
+@require_http_methods(["GET", "POST"])
+def mcp_token_create(request: HttpRequest) -> HttpResponse:
+    if request.method == "POST":
+        form = MCPTokenCreateForm(request.POST)
+        if form.is_valid():
+            token, plaintext = MCPToken.objects.create_token(
+                user=request.user,
+                name=form.cleaned_data["name"],
+                allowed_tools=form.cleaned_data.get("allowed_tools") or [],
+                expires_at=form.cleaned_data.get("expires_at") or None,
+            )
+            return render(
+                request,
+                "mcp_server/tokens/created.html",
+                {"token": token, "plaintext": plaintext},
+            )
+    else:
+        form = MCPTokenCreateForm()
+
+    return render(request, "mcp_server/tokens/create.html", {"form": form})
+
+
+@login_required
+@require_GET
+def mcp_token_detail(request: HttpRequest, pk: int) -> HttpResponse:
+    token = get_object_or_404(MCPToken, pk=pk, user=request.user)
+    return render(request, "mcp_server/tokens/detail.html", {"token": token})
+
+
+@login_required
+@require_http_methods(["GET", "POST"])
+def mcp_token_edit(request: HttpRequest, pk: int) -> HttpResponse:
+    token = get_object_or_404(MCPToken, pk=pk, user=request.user)
+
+    if request.method == "POST":
+        form = MCPTokenEditForm(request.POST, instance=token)
+        if form.is_valid():
+            instance = form.save(commit=False)
+            instance.allowed_tools = form.cleaned_data.get("allowed_tools") or []
+            instance.save()
+            messages.success(request, "MCP token updated.")
+            return redirect("mcp_server:mcp-token-detail", pk=token.pk)
+    else:
+        form = MCPTokenEditForm(instance=token)
+
+    return render(
+        request, "mcp_server/tokens/edit.html", {"form": form, "token": token}
+    )
+
+
+@login_required
+@require_POST
+def mcp_token_revoke(request: HttpRequest, pk: int) -> HttpResponse:
+    token = get_object_or_404(MCPToken, pk=pk, user=request.user)
+    token.is_active = False
+    token.save(update_fields=["is_active", "updated_at"])
+    messages.success(request, f"Revoked “{token.name}”. The token can no longer be used.")
+    return redirect("mcp_server:mcp-token-detail", pk=token.pk)
+
+
+@login_required
+@require_POST
+def mcp_token_delete(request: HttpRequest, pk: int) -> HttpResponse:
+    token = get_object_or_404(MCPToken, pk=pk, user=request.user)
+    name = token.name
+    token.delete()
+    messages.success(request, f"Deleted “{name}”.")
+    return redirect("mcp_server:mcp-token-list")

mnemosyne/mnemosyne/asgi.py

@@ -20,7 +20,7 @@ django.setup()
 
 from django.core.asgi import get_asgi_application  # noqa: E402
 from starlette.applications import Starlette  # noqa: E402
-from starlette.responses import JSONResponse  # noqa: E402
+from starlette.responses import JSONResponse, RedirectResponse  # noqa: E402
 from starlette.routing import Mount, Route  # noqa: E402
 
 from mcp_server.server import mcp  # noqa: E402
@@ -35,6 +35,13 @@ async def health(request):
     return JSONResponse({"status": "ok"})
 
 
+async def mcp_redirect(request):
+    """Bare /mcp → /mcp/ — MCP clients that omit the trailing slash hit
+    Starlette's Mount which only matches /mcp/* paths, falling through to
+    Django and returning 404. A 307 preserves the POST body."""
+    return RedirectResponse(url="/mcp/", status_code=307)
+
+
 @asynccontextmanager
 async def lifespan(app):
     async with mcp_http_app.lifespan(app), mcp_sse_app.lifespan(app):
@@ -44,6 +51,7 @@ async def lifespan(app):
 app = Starlette(
     routes=[
         Route("/mcp/health", health),
+        Route("/mcp", mcp_redirect, methods=["GET", "POST"]),
         Mount("/mcp/sse", app=mcp_sse_app),
         Mount("/mcp", app=mcp_http_app),
         Mount("/", app=application),

mnemosyne/mnemosyne/settings.py

@@ -181,6 +181,18 @@ else:
         },
     }
 
+# --- Daedalus S3 (cross-bucket reads for ingest) ---
+# Mnemosyne ingests files written to Daedalus's S3 bucket. These vars
+# configure read access; the file is copied into AWS_STORAGE_BUCKET_NAME
+# (Mnemosyne's own bucket) by the Celery ingest task before processing.
+DAEDALUS_S3_ENDPOINT_URL = env("DAEDALUS_S3_ENDPOINT_URL", default="")
+DAEDALUS_S3_ACCESS_KEY_ID = env("DAEDALUS_S3_ACCESS_KEY_ID", default="")
+DAEDALUS_S3_SECRET_ACCESS_KEY = env("DAEDALUS_S3_SECRET_ACCESS_KEY", default="")
+DAEDALUS_S3_BUCKET_NAME = env("DAEDALUS_S3_BUCKET_NAME", default="daedalus")
+DAEDALUS_S3_REGION_NAME = env("DAEDALUS_S3_REGION_NAME", default="us-east-1")
+DAEDALUS_S3_USE_SSL = env.bool("DAEDALUS_S3_USE_SSL", default=False)
+DAEDALUS_S3_VERIFY = env.bool("DAEDALUS_S3_VERIFY", default=True)
+
 # --- Celery / RabbitMQ ---
 CELERY_BROKER_URL = env(
     "CELERY_BROKER_URL",
@@ -196,6 +208,7 @@ CELERY_TASK_ACKS_LATE = True
 CELERY_WORKER_PREFETCH_MULTIPLIER = 1
 CELERY_TASK_ROUTES = {
     "library.tasks.embed_*": {"queue": "embedding"},
+    "library.tasks.ingest_*": {"queue": "embedding"},
     "library.tasks.batch_*": {"queue": "batch"},
 }
 

mnemosyne/mnemosyne/urls.py

@@ -25,4 +25,6 @@ urlpatterns = [
     path("library/", include("library.urls")),
     # LLM Manager
     path("llm/", include("llm_manager.urls")),
+    # MCP server (token dashboard at /profile/mcp-tokens/)
+    path("", include("mcp_server.urls")),
 ]

mnemosyne/test_db_manager/django_integration.py

@@ -52,15 +52,20 @@ class PostgreSQLTestRunner(DiscoverRunner):
         from django.db import connections
 
         db_cfg = self.pg_manager.get_django_database_config()
-        # Preserve Django's defaulted TEST sub-dict (CHARSET/MIRROR/MIGRATE…).
-        existing_test = connections["default"].settings_dict.get("TEST", {})
-        merged_test = {**existing_test, **db_cfg.get("TEST", {})}
-        db_cfg["TEST"] = merged_test
-        settings.DATABASES["default"] = db_cfg
+        # Preserve Django's defaulted top-level keys (ATOMIC_REQUESTS,
+        # AUTOCOMMIT, OPTIONS, …) and the TEST sub-dict (CHARSET, MIRROR,
+        # MIGRATE, …) — these are populated lazily by Django and absent from
+        # the user's raw settings.DATABASES, so a naive overwrite breaks
+        # request handling that consults them.
+        existing = connections["default"].settings_dict
+        existing_test = existing.get("TEST", {})
+        merged = {**existing, **db_cfg}
+        merged["TEST"] = {**existing_test, **db_cfg.get("TEST", {})}
+        settings.DATABASES["default"] = merged
         # The default connection was instantiated at Django bootstrap; its
-        # settings_dict is independent of settings.DATABASES.  Sync it
+        # settings_dict is independent of settings.DATABASES. Sync it
         # manually so test code talks to the container, not the dev DB.
-        connections["default"].settings_dict.update(db_cfg)
+        connections["default"].settings_dict.update(merged)
         logger.info("PostgreSQL test DB ready on port %s", self.pg_manager.assigned_port)
 
         # ── Neo4j ──────────────────────────────────────────────────────

mnemosyne/themis/migrations/0004_alter_userapikey_key_type.py

@@ -0,0 +1,18 @@
+# Generated by Django 5.2.13 on 2026-04-27 11:30
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('themis', '0003_alter_userprofile_current_timezone_and_more'),
+    ]
+
+    operations = [
+        migrations.AlterField(
+            model_name='userapikey',
+            name='key_type',
+            field=models.CharField(choices=[('api', 'API Key'), ('dav', 'DAV Credentials'), ('token', 'Access Token'), ('secret', 'Secret Key'), ('other', 'Other')], default='api', help_text='Type of credential', max_length=30),
+        ),
+    ]

mnemosyne/themis/models.py

@@ -228,7 +228,6 @@ class UserAPIKey(models.Model):
 
     KEY_TYPE_CHOICES = [
         ("api", "API Key"),
-        ("mcp", "MCP Server"),
         ("dav", "DAV Credentials"),
         ("token", "Access Token"),
         ("secret", "Secret Key"),

mnemosyne/themis/templates/themis/includes/user_menu.html

@@ -38,6 +38,16 @@
                 API Keys
             </a>
         </li>
+        <li>
+            <a href="{% url 'mcp_server:mcp-token-list' %}">
+                <svg xmlns="http://www.w3.org/2000/svg" class="h-4 w-4" fill="none"
+                     viewBox="0 0 24 24" stroke="currentColor">
+                    <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2"
+                          d="M9 12l2 2 4-4m5.618-4.016A11.955 11.955 0 0112 2.944a11.955 11.955 0 01-8.618 3.04A12.02 12.02 0 003 9c0 5.591 3.824 10.29 9 11.622 5.176-1.332 9-6.03 9-11.622 0-1.042-.133-2.052-.382-3.016z" />
+                </svg>
+                MCP Tokens
+            </a>
+        </li>
         <div class="divider my-0"></div>
         <li>
             <form method="post" action="{% url 'logout' %}">

mnemosyne/themis/tests/test_forms.py

@@ -248,7 +248,7 @@ class APIKeyEditFormTest(TestCase):
         form = APIKeyEditForm(
             data={
                 "service_name": "Updated Service",
-                "key_type": "mcp",
+                "key_type": "token",
                 "label": "Updated",
                 "instructions": "",
                 "help_url": "",
@@ -260,6 +260,6 @@ class APIKeyEditFormTest(TestCase):
         form.save()
         self.key.refresh_from_db()
         self.assertEqual(self.key.service_name, "Updated Service")
-        self.assertEqual(self.key.key_type, "mcp")
+        self.assertEqual(self.key.key_type, "token")
         self.assertFalse(self.key.is_active)
         self.assertEqual(self.key.encrypted_value, original_encrypted)

mnemosyne/themis/tests/test_models.py

@@ -209,7 +209,6 @@ class UserAPIKeyModelTest(TestCase):
         """KEY_TYPE_CHOICES contains expected types."""
         type_keys = [t[0] for t in UserAPIKey.KEY_TYPE_CHOICES]
         self.assertIn("api", type_keys)
-        self.assertIn("mcp", type_keys)
         self.assertIn("dav", type_keys)
         self.assertIn("token", type_keys)
         self.assertIn("secret", type_keys)

nginx/mnemosyne.conf

@@ -0,0 +1,86 @@
+# Mnemosyne nginx — single virtual host that fronts the Django web app
+# and the FastMCP server. HAProxy on Titania terminates TLS and routes by
+# hostname; this nginx is plain HTTP on the internal network.
+
+# Map of upstreams to give us readable proxy_pass targets and easy retries.
+upstream mnemosyne_web {
+    server web:8000 max_fails=3 fail_timeout=30s;
+}
+
+upstream mnemosyne_mcp {
+    server mcp:22091 max_fails=3 fail_timeout=30s;
+}
+
+server {
+    listen 80 default_server;
+    server_name _;
+
+    # Reasonable limits — file uploads to the ingest endpoint can be big,
+    # but the bulk path is S3-direct from Daedalus. 64 MB covers admin
+    # uploads and direct REST POST /library/api/items/upload.
+    client_max_body_size 64m;
+    client_body_timeout  120s;
+
+    # Mnemosyne's REST API — Django REST Framework views + admin.
+    # Under /library/api/* per mnemosyne/urls.py and /admin/* per Django.
+    location /library/ {
+        proxy_pass http://mnemosyne_web;
+        proxy_http_version 1.1;
+        proxy_set_header Host              $host;
+        proxy_set_header X-Real-IP         $remote_addr;
+        proxy_set_header X-Forwarded-For   $proxy_add_x_forwarded_for;
+        proxy_set_header X-Forwarded-Proto $scheme;
+        proxy_read_timeout 300s;
+    }
+
+    location /admin/ {
+        proxy_pass http://mnemosyne_web;
+        proxy_http_version 1.1;
+        proxy_set_header Host              $host;
+        proxy_set_header X-Real-IP         $remote_addr;
+        proxy_set_header X-Forwarded-For   $proxy_add_x_forwarded_for;
+        proxy_set_header X-Forwarded-Proto $scheme;
+        proxy_read_timeout 300s;
+    }
+
+    # FastMCP Streamable HTTP at /mcp/ and SSE at /mcp/sse/.
+    # Long-running streams need disabled buffering and a generous timeout.
+    location /mcp/ {
+        proxy_pass http://mnemosyne_mcp;
+        proxy_http_version 1.1;
+        proxy_set_header Host              $host;
+        proxy_set_header X-Real-IP         $remote_addr;
+        proxy_set_header X-Forwarded-For   $proxy_add_x_forwarded_for;
+        proxy_set_header X-Forwarded-Proto $scheme;
+        proxy_set_header Connection        "";
+        proxy_buffering    off;
+        proxy_cache        off;
+        proxy_read_timeout 600s;
+    }
+
+    # Static files baked into the image at /app/staticfiles, mounted into
+    # this nginx via a named volume populated by the web service.
+    location /static/ {
+        alias /var/www/static/;
+        access_log off;
+        expires 30d;
+    }
+
+    # Prometheus scrape endpoint — internal networks only.
+    # Allows: localhost + RFC1918 private ranges (10/8, 172.16/12, 192.168/16).
+    location /metrics {
+        allow 127.0.0.0/8;
+        allow 10.0.0.0/8;
+        allow 172.16.0.0/12;
+        allow 192.168.0.0/16;
+        deny  all;
+        proxy_pass http://mnemosyne_web;
+        access_log off;
+    }
+
+    # Liveness probe — proxies through to the MCP health endpoint.
+    location = /healthz {
+        proxy_pass http://mnemosyne_mcp/mcp/health;
+        access_log off;
+    }
+}

pyproject.toml

@@ -40,6 +40,17 @@ dev = [
     "django-debug-toolbar>=4.0,<5.0",
     "docker>=7.0,<8.0",
 ]
+test = [
+    "pytest>=8.0,<9.0",
+    "pytest-django>=4.8,<5.0",
+]
+lint = [
+    "ruff>=0.6,<1.0",
+]
+docs = [
+    "mkdocs>=1.6,<2.0",
+    "mkdocs-material>=9.5,<10.0",
+]
 
 [build-system]
 requires = ["setuptools>=68.0"]

validator/.gitignore

@@ -0,0 +1,6 @@
+fastagent.secrets.yaml
+fastagent.jsonl
+.venv/
+*.egg-info/
+__pycache__/
+*.pyc

validator/README.md

@@ -0,0 +1,101 @@
+# Mnemosyne Validator
+
+A bare [FastAgent](https://github.com/evalstate/fast-agent) + [Pallas](https://git.helu.ca/r/pallas) project whose only purpose is to exercise Mnemosyne's MCP server end-to-end. Use it to confirm the transport works, every MCP tool registers, args/responses round-trip, and the local LLM can actually drive the tools.
+
+This is **not** a production agent. It does not represent the long-term Daedalus integration — when Daedalus ships, it will inject `workspace_id` server-side. The validator never sets `workspace_id`, meaning all calls run against the global scope (libraries with `workspace_id IS NULL`).
+
+## Layout
+
+```
+validator/
+├── pyproject.toml                 # pallas-mcp + fast-agent-mcp deps
+├── agents.yaml                    # one-agent Pallas topology
+├── fastagent.config.yaml          # default_model + mnemosyne MCP server
+├── fastagent.secrets.yaml.example # schema for the (gitignored) secrets file
+└── agents/
+    └── mnemosyne_validator.py     # the FastAgent definition
+```
+
+All configuration lives in the FastAgent YAML files — no `.env` is used. The
+real `fastagent.secrets.yaml` is gitignored; `fastagent.secrets.yaml.example`
+documents its shape and is checked in.
+
+## Setup
+
+```bash
+cd validator/
+python3.13 -m venv .venv
+source .venv/bin/activate
+pip install -e .
+```
+
+Copy the secrets template and fill in the bearer token (see next section):
+
+```bash
+cp fastagent.secrets.yaml.example fastagent.secrets.yaml
+```
+
+## Provision an MCP bearer token
+
+Mnemosyne requires a bearer token when `MCP_REQUIRE_AUTH=True` (the default). Generate one for your user:
+
+```bash
+cd ../mnemosyne
+python manage.py create_mcp_token --user <username> --name validator
+```
+
+The command prints the token **once** — paste it into `validator/fastagent.secrets.yaml` under `mcp.servers.mnemosyne.headers.Authorization` (keep the `Bearer ` prefix).
+
+## Start Mnemosyne's MCP server
+
+The validator hits the Mnemosyne ASGI endpoint, so Mnemosyne's MCP server must be running. From the Mnemosyne project:
+
+```bash
+cd mnemosyne/
+uvicorn mnemosyne.asgi:app --host 0.0.0.0 --port 22091 --workers 1
+```
+
+By default the validator points at `http://localhost:22091/mcp/` (note the trailing slash — Starlette's `Mount` only matches paths under the mount prefix, so `/mcp` without the slash falls through to Django). If your Mnemosyne is on another host, override `mcp.servers.mnemosyne.url` in `fastagent.secrets.yaml`.
+
+## Run the validator
+
+The `mnemosyne-validator` script is a thin alias for `pallas`:
+
+```bash
+# Start with the registry (Pallas mode):
+mnemosyne-validator
+
+# Or run the agent directly (no registry):
+mnemosyne-validator --agent mnemosyne_validator
+```
+
+To chat with the agent directly without spinning up a Pallas registry, use the `fast-agent` CLI (provided by `fast-agent-mcp`):
+
+```bash
+fast-agent go --config-path fastagent.config.yaml --url http://localhost:24301/mcp mnemosyne_validator
+```
+
+## What to test
+
+These prompts exercise every Mnemosyne MCP tool. After each, the agent should call the named tool and surface the result.
+
+| Prompt | Tool | What to verify |
+|--------|------|----------------|
+| "Run a health check on Mnemosyne." | `get_health` | Returns `status: ok` if Neo4j + S3 + embedding model are all reachable. `degraded` if one is down. |
+| "List all libraries." | `list_libraries` | Returns the libraries seeded by `load_library_types`, each with `library_type` set. |
+| "List collections in library `<uid>`." | `list_collections` | Returns collections inside the named library. |
+| "List items in collection `<uid>`." | `list_items` | Returns items with `chunk_count` and `embedding_status`. |
+| "Search the technical libraries for `<query>`." | `search` | Returns ranked candidates with `chunk_uid`, `score`, `text_preview`, `library_type`. |
+| "Fetch the full text of chunk `<chunk_uid>`." | `get_chunk` | Returns the full chunk text from S3. |
+
+If a call errors, the agent surfaces it verbatim — that's the failure mode you want.
+
+## Troubleshooting
+
+**"Invalid MCP token"** — token wasn't provisioned, was provisioned for a different user, or got mangled when pasted. Re-run `create_mcp_token` and paste again. Tokens are SHA-256 hashed at rest and can't be retrieved later.
+
+**"Couldn't connect to Mnemosyne"** — the ASGI server isn't running, or it's bound to a different host/port than `mcp.servers.mnemosyne.url` says. Check `curl http://localhost:22091/mcp/health` returns `{"status":"ok"}`.
+
+**"No system embedding model configured" in `get_health`** — `LLMModel.get_system_embedding_model()` returns nothing. Configure the embedding model via the Mnemosyne admin or `manage.py` before searches will work.
+
+**Search returns zero candidates with no error** — Mnemosyne is reachable but has no embedded content yet. Upload an item and run `embed_item`, or use the Daedalus ingest endpoint, before re-testing search.

validator/agents.yaml

@@ -0,0 +1,17 @@
+# Mnemosyne Validator — Pallas deployment topology
+#
+# A single-agent Pallas project whose only purpose is to validate the
+# Mnemosyne MCP server end-to-end. Not a production deployment.
+
+name: mnemosyne-validator
+version: "0.1.0"
+host: localhost
+namespace: ca.helu.mnemosyne-validator
+registry_port: 24300
+
+agents:
+  mnemosyne_validator:
+    module: agents.mnemosyne_validator
+    port: 24301
+    title: Mnemosyne Validator
+    description: "Exercises Mnemosyne's MCP tools: search, get_chunk, list_*, get_health"

validator/agents/mnemosyne_validator.py

@@ -0,0 +1,56 @@
+"""
+Mnemosyne Validator Agent
+
+A bare FastAgent that wraps the Mnemosyne MCP server. Exists solely to
+exercise the Mnemosyne MCP transport, tool registration, and round-trip
+serialization — no production role.
+
+Drive it from the CLI to confirm:
+- search works against the running Mnemosyne (vector + fulltext + graph)
+- get_chunk fetches full chunk text from S3
+- list_libraries / list_collections / list_items return the expected shape
+- get_health returns ok/degraded with the right dependency breakdown
+
+When the Daedalus integration ships, the workspace_id parameter will be
+injected by Daedalus's chat path (force-overwritten before the call leaves
+Daedalus). This validator never sets it — meaning all calls go to the
+GLOBAL scope (libraries with workspace_id IS NULL).
+"""
+from fast_agent import FastAgent
+
+fast = FastAgent("Mnemosyne Validator", parse_cli_args=False)
+
+
+@fast.agent(
+    name="mnemosyne_validator",
+    instruction="""You are a validator for the Mnemosyne knowledge base. Your
+job is to exercise its MCP tools when asked, report what you saw, and surface
+errors clearly.
+
+You have direct access to Mnemosyne via these tools:
+
+- search(query, library_uid?, library_type?, collection_uid?, limit?, rerank?, include_images?, search_types?)
+    Hybrid retrieval. Returns ranked chunks with text_preview (~500 chars),
+    chunk_uid, item_uid, item_title, library_type, score, source.
+- get_chunk(chunk_uid)
+    Fetch the full text of a chunk by uid (typically obtained from search).
+- list_libraries(limit?, offset?)
+    List libraries (uid, name, library_type, description).
+- list_collections(library_uid?, limit?, offset?)
+    List collections, optionally filtered by parent library.
+- list_items(collection_uid?, library_uid?, limit?, offset?)
+    List items (documents) with chunk_count, embedding_status, etc.
+- get_health()
+    Health check: {status: ok|degraded|error, checks: {neo4j, s3, embedding}}.
+
+When the user asks "what libraries exist", call list_libraries and report.
+When they ask a research question, call search and surface chunk_uid + score
++ item_title for each candidate. If they want full text, call get_chunk.
+Show raw structured output, not flowery prose — this is a validation tool,
+not a chat assistant.
+
+If a tool errors, paste the error message verbatim.""",
+    servers=["mnemosyne"],
+)
+async def mnemosyne_validator():
+    pass

validator/fastagent.config.yaml

@@ -0,0 +1,32 @@
+# Mnemosyne Validator — FastAgent + MCP configuration
+#
+# Secrets (api_key, MCP bearer tokens) live in fastagent.secrets.yaml
+# (gitignored) and merge with this file at runtime.
+
+# Local llama.cpp on Nyx (OpenAI-compatible). Override via
+# fastagent.secrets.yaml if you want to point at a different model server.
+default_model: openai.Qwen3.5-35B-A3B-UD-Q4_K_XL.gguf
+
+# Capabilities for the model — Pallas registers it with fast-agent's
+# ModelDatabase using these values. vision: true so we can validate image
+# round-trip later (search returns image candidates by default).
+model_capabilities:
+  vision: true
+  context_window: 192000
+  max_output_tokens: 16384
+
+# ── LLM Providers ───────────────────────────────────────────────────────────
+openai:
+  base_url: "http://nyx.helu.ca:22079/v1"
+
+# ── MCP Servers ─────────────────────────────────────────────────────────────
+mcp:
+  servers:
+    # Mnemosyne MCP server — Streamable HTTP at /mcp.
+    # Default assumes the validator runs on the same host as Mnemosyne;
+    # override the URL in fastagent.secrets.yaml or via Ansible if remote.
+    mnemosyne:
+      transport: http
+      url: "http://localhost:22091/mcp/"
+      # Bearer token in fastagent.secrets.yaml (provisioned via
+      # `python manage.py create_mcp_token <user>`).

validator/fastagent.secrets.yaml.example

@@ -0,0 +1,22 @@
+# Mnemosyne Validator — secrets template
+#
+# Copy to fastagent.secrets.yaml and fill in real values. The .yaml is
+# gitignored; the .yaml.example is committed.
+
+# ── LLM provider keys ───────────────────────────────────────────────────────
+# Local llama.cpp doesn't authenticate, but fast-agent requires the key field
+# to be present. "0000" or any non-empty string is fine.
+openai:
+  api_key: "0000"
+
+# ── MCP server bearer tokens ────────────────────────────────────────────────
+mcp:
+  servers:
+    mnemosyne:
+      headers:
+        # Mnemosyne MCP server requires a bearer token when MCP_REQUIRE_AUTH=True.
+        # Provision one with:
+        #     cd ../mnemosyne
+        #     python manage.py create_mcp_token --user <username> --name validator
+        # then paste the printed token here (it is shown once and not retrievable).
+        Authorization: "Bearer paste-mcp-token-here"

validator/pyproject.toml

@@ -0,0 +1,23 @@
+[project]
+name = "mnemosyne-validator"
+version = "0.1.0"
+description = "FastAgent + Pallas validator that talks to Mnemosyne's MCP server end-to-end"
+requires-python = ">=3.13"
+dependencies = [
+    "pallas-mcp @ git+ssh://git@git.helu.ca:22022/r/pallas.git",
+    "fast-agent-mcp>=0.6.10",
+    "pyyaml>=6.0",
+]
+
+[project.scripts]
+mnemosyne-validator = "pallas.server:main"
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.metadata]
+allow-direct-references = true
+
+[tool.hatch.build.targets.wheel]
+packages = ["agents"]