AI coding agents can read your source code but not your documentation. They guess at architecture, miss established patterns, and hallucinate details they could have looked up.

Gnosis MCP fixes this. Point it at a folder of markdown files and it creates a searchable knowledge base that any MCP-compatible AI agent can query — Claude Code, Cursor, Windsurf, Cline, and any tool that supports the Model Context Protocol.

No database server. SQLite works out of the box with keyword search, or add [embeddings] for local semantic search. Scale to PostgreSQL + pgvector when needed.

Why use this

Less hallucination. Agents search your docs before guessing. Architecture decisions, API contracts, billing rules — one tool call away instead of made up.

Lower token costs. A search returns ~600 tokens of ranked results. Reading the same docs as files costs 3,000-8,000+ tokens. On a 170-doc knowledge base (~840K tokens), that's the difference between a precise answer and a blown context window.

Docs that stay current. Add a new markdown file, run ingest, it's searchable immediately. Or use --watch to auto-re-ingest on file changes. No routing tables to maintain, no hardcoded paths to update.

Works with what you have. Your docs are already markdown files in a folder. Gnosis MCP indexes them as-is — no format conversion, no special syntax needed.

Quick Start

pip install gnosis-mcp
gnosis-mcp ingest ./docs/       # loads markdown, auto-creates SQLite database
gnosis-mcp serve                # starts MCP server

That's it. Your AI agent can now search your docs.

Want semantic search? Add local ONNX embeddings (no API key needed, ~23MB model):

pip install gnosis-mcp[embeddings]
gnosis-mcp ingest ./docs/ --embed   # ingest + embed in one step
gnosis-mcp serve                    # hybrid keyword+semantic search auto-activated

Test it before connecting to an editor:

gnosis-mcp search "getting started"           # keyword search
gnosis-mcp search "how does auth work" --embed # hybrid semantic+keyword
gnosis-mcp stats                               # see what was indexed

uvx gnosis-mcp ingest ./docs/
uvx gnosis-mcp serve

Editor Integrations

Gnosis MCP works with any MCP-compatible editor. Add the server config, and your AI agent gets search_docs, get_doc, and get_related tools automatically.

Claude Code

Add to .claude/mcp.json:

{
  "mcpServers": {
    "docs": {
      "command": "gnosis-mcp",
      "args": ["serve"]
    }
  }
}

Or install as a Claude Code plugin for a richer experience with slash commands.

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "docs": {
      "command": "gnosis-mcp",
      "args": ["serve"]
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "docs": {
      "command": "gnosis-mcp",
      "args": ["serve"]
    }
  }
}

VS Code (GitHub Copilot)

Add to .vscode/mcp.json in your workspace:

{
  "servers": {
    "docs": {
      "command": "gnosis-mcp",
      "args": ["serve"]
    }
  }
}

Also discoverable via the VS Code MCP gallery — search @mcp gnosis in the Extensions view.

Enterprise: Your org admin needs the "MCP servers in Copilot" policy enabled. Free/Pro/Pro+ plans work without this.

JetBrains (IntelliJ, PyCharm, WebStorm)

Go to Settings > Tools > AI Assistant > MCP Servers, click +, and add:

• Name: docs
• Command: gnosis-mcp
• Arguments: serve

Cline

Open Cline MCP settings panel and add the same server config.

Other MCP clients

Any tool that supports the Model Context Protocol works — including Zed, Neovim (via plugins), and custom agents. The server communicates over stdio by default, or SSE with --transport sse.

Choose Your Backend

Auto-detection: Set DATABASE_URL to postgresql://... and it uses PostgreSQL. Don't set it and it uses SQLite. Override with GNOSIS_MCP_BACKEND=sqlite|postgres.

pip install gnosis-mcp[postgres]
export GNOSIS_MCP_DATABASE_URL="postgresql://user:pass@localhost:5432/mydb"
gnosis-mcp init-db              # create tables + indexes
gnosis-mcp ingest ./docs/       # load your markdown
gnosis-mcp serve

CREATE EXTENSION IF NOT EXISTS vector;

gnosis-mcp embed                        # via OpenAI (default)
gnosis-mcp embed --provider ollama      # or use local Ollama

Claude Code Plugin

For Claude Code users, install as a plugin to get the MCP server plus slash commands:

claude plugin marketplace add nicholasglazer/gnosis-mcp
claude plugin install gnosis

This gives you:

The plugin works with both SQLite and PostgreSQL backends.

{
  "mcpServers": {
    "gnosis": {
      "command": "gnosis-mcp",
      "args": ["serve"]
    }
  }
}

What It Does

Gnosis MCP exposes 6 tools and 3 resources over MCP. Your AI agent calls these automatically when it needs information from your docs.

Tools

Read tools are always available. Write tools require GNOSIS_MCP_WRITABLE=true.

Resources

How search works

# Keyword search — works on both SQLite and PostgreSQL
gnosis-mcp search "stripe webhook"

# Hybrid search — keyword + semantic similarity (PostgreSQL + embeddings)
gnosis-mcp search "how does billing work" --embed

# Filtered — narrow results to a specific category
gnosis-mcp search "auth" -c guides

When called via MCP, the agent passes a query string for keyword search. On PostgreSQL with embeddings, it can also pass query_embedding for hybrid mode that combines keyword matching with semantic similarity.

Search results include a highlight field with matched terms wrapped in <mark> tags for context-aware snippets (FTS5 snippet() on SQLite, ts_headline() on PostgreSQL).

Embeddings

Embeddings enable semantic search — finding docs by meaning, not just keywords.

1. Local ONNX (recommended for SQLite) — zero-config, no API key needed:

pip install gnosis-mcp[embeddings]
gnosis-mcp ingest ./docs/ --embed       # ingest + embed in one step
gnosis-mcp embed                        # or embed existing chunks separately

Uses MongoDB/mdbr-leaf-ir (~23MB quantized, Apache 2.0). Auto-downloads on first run. Customize with GNOSIS_MCP_EMBED_MODEL.

2. Remote providers — OpenAI, Ollama, or any OpenAI-compatible endpoint:

gnosis-mcp embed --provider openai      # requires GNOSIS_MCP_EMBED_API_KEY
gnosis-mcp embed --provider ollama      # uses local Ollama server

3. Pre-computed vectors — pass embeddings to upsert_doc or query_embedding to search_docs from your own pipeline.

Hybrid search — when embeddings are available, search automatically combines keyword (BM25) and semantic (cosine) results using Reciprocal Rank Fusion (RRF). Works on both SQLite (via sqlite-vec) and PostgreSQL (via pgvector).

Configuration

All settings via environment variables. Nothing required for SQLite — it works with zero config.

CREATE FUNCTION my_schema.my_search(
    p_query_text text,
    p_categories text[],
    p_limit integer
) RETURNS TABLE (
    file_path text, title text, content text,
    category text, combined_score double precision
) ...

GNOSIS_MCP_SEARCH_FUNCTION=my_schema.my_search

GNOSIS_MCP_CHUNKS_TABLE=documentation_chunks,api_docs,tutorial_chunks

CLI Reference

gnosis-mcp ingest <path> [--dry-run] [--embed]             Load markdown files (--embed to generate embeddings)
gnosis-mcp serve [--transport stdio|sse] [--ingest PATH] [--watch PATH]   Start MCP server (--watch for live reload)
gnosis-mcp search <query> [-n LIMIT] [-c CAT] [--embed]    Search (--embed for hybrid semantic+keyword)
gnosis-mcp stats                                           Show document, chunk, and embedding counts
gnosis-mcp check                                           Verify database connection + sqlite-vec status
gnosis-mcp embed [--provider P] [--model M] [--dry-run]    Backfill embeddings (auto-detects local provider)
gnosis-mcp init-db [--dry-run]                             Create tables + indexes manually
gnosis-mcp export [-f json|markdown] [-c CAT]              Export documents

How ingestion works

gnosis-mcp ingest scans a directory for .md files and loads them into the database:

• Smart chunking — splits by H2 headings, keeping sections together (not arbitrary character limits)
• Frontmatter support — extracts title, category, audience, tags from YAML frontmatter
• Auto-linking — relates_to in frontmatter creates bidirectional links (queryable via get_related)
• Auto-categorization — infers category from the parent directory name
• Incremental updates — content hashing skips unchanged files on re-run
• Watch mode — gnosis-mcp serve --watch ./docs/ auto-re-ingests on file changes
• Dry run — preview what would be indexed with --dry-run

Available on

Gnosis MCP is listed on the Official MCP Registry (which feeds the VS Code MCP gallery and GitHub Copilot), PyPI, and major MCP directories including mcp.so, Glama, and cursor.directory.

Architecture

src/gnosis_mcp/
├── backend.py         DocBackend protocol + create_backend() factory
├── pg_backend.py      PostgreSQL — asyncpg, tsvector, pgvector
├── sqlite_backend.py  SQLite — aiosqlite, FTS5, sqlite-vec hybrid search (RRF)
├── sqlite_schema.py   SQLite DDL — tables, FTS5, triggers, vec0 virtual table
├── config.py          Config from env vars, backend auto-detection
├── db.py              Backend lifecycle + FastMCP lifespan
├── server.py          FastMCP server — 6 tools, 3 resources, auto-embed queries
├── ingest.py          Markdown scanner — H2 chunking, frontmatter
├── watch.py           File watcher — mtime polling, auto-re-ingest on changes
├── schema.py          PostgreSQL DDL — tables, indexes, search functions
├── embed.py           Embedding providers — OpenAI, Ollama, custom, local ONNX
├── local_embed.py     Local ONNX embedding engine — HuggingFace model download
└── cli.py             CLI — serve, ingest, search, embed, stats, check

AI-Friendly Docs

These files are optimized for AI agents to consume:

Development

git clone https://github.com/nicholasglazer/gnosis-mcp.git
cd gnosis-mcp
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest                    # 220+ tests, no database needed
ruff check src/ tests/

All tests run without a database. Keep it that way.

Good first contributions: new embedding providers, export formats, ingestion for RST/AsciiDoc/HTML, search highlighting. Open an issue first for larger changes.

Sponsors

If Gnosis MCP saves you time, consider sponsoring the project.

License

MIT