mnemo-brain-mcp

MCP (Model Context Protocol) bridge to query your Mnemo second brain from Claude Code, Cursor, Cline, Windsurf, and any MCP-compatible AI tool.

What is this?

This is a friendly fork of desimpkins/daniel-lightrag-mcp. The protocol logic is identical; the package is renamed for one reason: a clean PyPI install for Mnemo users.

If you're using a standalone LightRAG server (not Mnemo), the upstream package works fine — and you'll get upstream's updates faster.

Install

pip install mnemo-brain-mcp

After install the mnemo-brain-mcp command is on your PATH.

Configure

Point the bridge at your local LightRAG-API server (Mnemo runs it at 127.0.0.1:9621 by default):

Claude Code

Edit ~/.claude/claude_mcp_config.json (or run claude mcp add):

{
  "mcpServers": {
    "mnemo-brain": {
      "command": "mnemo-brain-mcp",
      "env": {
        "LIGHTRAG_BASE_URL": "http://localhost:9621",
        "LIGHTRAG_API_KEY": "<paste contents of secrets/lightrag_api_key.txt>"
      }
    }
  }
}

Restart Claude Code. Try:

"What does my brain know about my current projects?"

Cursor

Settings → MCP → Add new MCP server. Same JSON shape as above.

Cline / Windsurf

See their docs for the mcpServers config location — the JSON shape is the same.

What this gives you

22 tools over MCP, including:

For the full tool list see the upstream docs in the archive/ directory.

Security note

This bridge exposes write tools (insert_text, update_entity, delete_document, ...). Granting your coding tool access to this MCP server grants it write access to the LightRAG knowledge graph.

For Mnemo specifically: the canonical writer to your Obsidian vault is the Telegram bot. Anything an MCP client insert_text's into the graph lives only in the LightRAG index, not in your Markdown notes. To write to your vault, message the bot.

Credit

Original code by @desimpkins. Renamed and republished with thanks. MIT-licensed both upstream and here.

License

MIT.