mnemosyne-mcp-server

Standalone Model Context Protocol (MCP) server for the Mnemosyne memory substrate. Exposes search, store, delete, context, list_projects, and get_profile tools to Claude Desktop, Cursor, and any MCP-compatible client.

Install

pip install mnemosyne-mcp-server

Configure

Set your API key:

export MNEMOSYNE_API_KEY="mn_..."
# Optional:
export MNEMOSYNE_BASE_URL="https://api.mnemosyne.geasslabs.xyz"
export MNEMOSYNE_DEFAULT_CONTAINER="default"

Run

stdio (default - for local clients like Claude Desktop)

mnemosyne-mcp

SSE / HTTP (for remote clients)

mnemosyne-mcp --transport sse --host 0.0.0.0 --port 8765

Use with Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "mnemosyne": {
      "command": "mnemosyne-mcp",
      "env": {
        "MNEMOSYNE_API_KEY": "mn_..."
      }
    }
  }
}

Restart Claude Desktop. You'll see the Mnemosyne tools (memory, recall, context, list_projects, get_profile, delete) in the tool menu.

Use with Cursor

In .cursor/mcp.json:

{
  "mcpServers": {
    "mnemosyne": {
      "command": "mnemosyne-mcp",
      "env": {
        "MNEMOSYNE_API_KEY": "mn_..."
      }
    }
  }
}

Available tools

For the full 29-tool surface (including dream, reflect, introspect, find_analogies, predict_outcome, what_if, etc.), use the in-process MCP server from the Mnemosyne monorepo.

License

MIT © Geass Labs