Gnosis MCP

Serve your PostgreSQL docs to AI agents over MCP.

Gnosis MCP demo — init, ingest, search, check
PyPI GitHub MIT License 2 dependencies

Quick Start

pip install gnosis-mcp
export GNOSIS_MCP_DATABASE_URL="postgresql://user:pass@localhost:5432/mydb"
gnosis-mcp init-db
gnosis-mcp ingest ./docs/
gnosis-mcp serve

What It Does

  • 6 MCP tools — search, get, related, upsert, delete, update metadata
  • 3 resources — document listing, content retrieval, category browsing
  • Hybrid search — keyword (tsvector) + semantic (pgvector cosine) with RRF scoring
  • Markdown ingestion — chunks by H2, extracts frontmatter, content hashing for fast re-ingestion
  • Embedding support — OpenAI, Ollama, or any compatible API (zero new deps)
  • Multi-table queries — serve docs from multiple tables via UNION ALL
  • Webhooks — get notified on doc changes
  • 2 dependencies — mcp + asyncpg, nothing else

Works With

Claude Code, Cursor, Windsurf, VS Code, Cline, and any MCP-compatible client.

Embeddings

Three tiers of embedding integration, all with zero new dependencies:

  • Pre-computed — pass embeddings directly via MCP tools
  • Backfill — gnosis-mcp embed fills NULL embeddings via OpenAI, Ollama, or custom API
  • Hybrid search — automatically combines keyword + semantic scoring when embeddings exist
GitHub PyPI AI Reference