Local Vector Store & MCP Server

Lightweight vector store with TF‑IDF search, a small FastAPI HTTP API, and an MCP stdio server. Ingests documents from input/html and input/PDF and stores artifacts under data/vector_store.

Usage (Local)

• Install: pip install -r requirements.txt
• Ingest data: make ingest (reads input/html and input/PDF)
• Query via CLI: make query Q="security maturity" K=5
• HTTP API (after deployment below):
  • Health: curl localhost:8000/health
  • Ingest: curl -X POST localhost:8000/ingest
  • Query: curl -X POST localhost:8000/query -H 'Content-Type: application/json' -d '{"query":"security maturity","k":5}'
• Vector Store Manager (interactive): make manage
  • Examples: status, docs --limit 10, chunks input/PDF/example.pdf --limit 5, search "zero trust" --k 5, ingest, purge, export assets/index_backup.jsonl, help, exit

Deployment

Docker (single container)

• Build: docker build -t local/vector-mcp:latest .
• Run: docker run -p 8000:8000 -e AUTO_INGEST=1 -v "$PWD/input:/app/input" -v "$PWD/data:/app/data" local/vector-mcp:latest
  • Visit http://localhost:8000/health or use curl examples above.

Docker Compose

• Build images: make docker-build
• Start services: make docker-up (HTTP server on :8000)
• View logs: make docker-logs
• Ingest inside container: make docker-ingest
• Query inside container: make docker-query Q="your query" K=5
• Stop: make docker-down

MCP Stdio Server

• Local: make mcp-stdio (runs python -m src.mcp_server)
• Compose service: make mcp-stdio-up (optional background service); make mcp-stdio-down to remove.

Data Layout

• Input: input/html/**/*.html, input/PDF/**/*.pdf
• Artifacts: data/vector_store/{vectorizer.json,index.jsonl,meta.json}

Notes

• Ensure input/ contains documents before running ingest.
• Set AUTO_INGEST=1 to ingest on container start (Docker only).