mnemo-mcp

MCP server for Mnemo — query your developer memory from any AI agent.

Mnemo captures your AI conversations (Claude Code, Cursor, ChatGPT, Gemini, Pi, claude.ai) and web clips into a local, searchable memory. This MCP server exposes that memory to any agent that speaks the Model Context Protocol.

Setup

Add to your MCP client configuration (Claude Desktop, Claude Code, Cursor, etc.):

{
  "mcpServers": {
    "mnemo": {
      "command": "npx",
      "args": ["-y", "@usemnemo/mnemo-mcp"]
    }
  }
}

Requirements

• Mnemo desktop app must be running for the MCP server to access your memory.
• The desktop app runs on localhost:47822 and handles all data storage, embeddings, and search.
• Node.js >= 18.

Available Tools

Search & retrieval

| Tool | Description | |---|---| | search_memory | Semantic + full-text search across all clips and AI conversations. Returns memory IDs with previews. | | get_memory | Fetch the full content of a memory by ID (use after search_memory). | | get_recent | Most recent captures across all sources — what you worked on lately. | | search_by_date | Search by date range, optionally filtered by source. |

Sessions & threads

| Tool | Description | |---|---| | list_sessions | List auto-clustered work sessions, optionally filtered by project wing. | | get_session | All memories and clips in a specific session. | | get_thread | Retrieve a full conversation thread by ID, with optional L2 summary. | | unfinished_threads | Conversation threads that never reached "resolved" status. |

Summaries & domains (L2/L3)

| Tool | Description | |---|---| | summarize_today | Coherent summary of everything worked on today, across all sources. | | list_domains | All knowledge domains discovered by Mnemo's L2 summarisation layer. | | memory_stats | Overall stats: memories, clips, sources, L1 extractions, L2 summaries, L3 fusions. |

Cognitive tools (Memeory consolidation)

| Tool | Description | |---|---| | context_recovery | Recover working context for a domain — decisions and open questions. | | tunnel_state | Current thinking stage for a domain (exploring/implementing/debugging/resolved). | | what_do_i_think | Aggregated past decisions and opinions on a topic. | | switching_cost | Estimated cognitive cost of switching to a different domain. | | dormant_contexts | Abandoned or dormant work — domains with unresolved questions inactive > 1 week. | | thinking_trajectory | How your thinking on a topic evolved over time. |

How It Works

AI Agent  ←→  mnemo-mcp (stdio)  ←→  Mnemo Desktop App (localhost:47822)  ←→  SQLite + Embeddings

The MCP server is a thin authenticated HTTP proxy. All search, embedding, summarisation, and storage happen in the desktop app. The MCP process fetches a handshake token on first request and retries once on 403 (desktop restart).

Development

npm install
npm run build
npm link        # makes 'mnemo-mcp' available globally

Publishing to npm

The package is published as @usemnemo/mnemo-mcp under the @usemnemo scope with public access.

npm view @usemnemo/mnemo-mcp version

License

MIT