@memory-module/mcp

MCP server for Memory Module — agentic memory for LLM agents.

Exposes the full Memory Module API as 10 MCP tools. Works with Claude Code, Cursor, Windsurf, VS Code Copilot, Zed, and any MCP-compatible client.

Quick Start

Add to your MCP client config:

Claude Code (.claude.json):

{
  "mcpServers": {
    "memory-module": {
      "command": "npx",
      "args": ["-y", "@memory-module/mcp"],
      "env": {
        "MEMORY_API_URL": "https://api.memorymodule.io",
        "MEMORY_API_KEY": "your-api-key"
      }
    }
  }
}

Cursor / Windsurf (.cursor/mcp.json or equivalent):

{
  "mcpServers": {
    "memory-module": {
      "command": "npx",
      "args": ["-y", "@memory-module/mcp"],
      "env": {
        "MEMORY_API_URL": "https://api.memorymodule.io",
        "MEMORY_API_KEY": "your-api-key"
      }
    }
  }
}

Tools

Memory

| Tool | Description | |------|-------------| | store_memory | Store a new memory with auto-generated keywords, tags, context, and links | | search_memories | Semantic search with similarity scores and linked memories | | list_memories | Paginated listing with optional dimension filters | | get_memory | Get a single memory by ID | | delete_memory | Delete a memory and clean up bidirectional links |

Batch & Ingestion

| Tool | Description | |------|-------------| | batch_store_memories | Store multiple memories asynchronously | | ingest_document | Chunk and store a long document as memories | | get_job_status | Check async job progress |

Configuration

| Tool | Description | |------|-------------| | list_dimensions | List registered custom filter dimensions | | register_dimensions | Register new filterable dimensions |

Configuration

| Variable | Required | Default | Description | |----------|----------|---------|-------------| | MEMORY_API_KEY | Yes | — | API key for Memory Module | | MEMORY_API_URL | No | http://localhost:8080 | Memory Module API base URL | | REQUEST_TIMEOUT_MS | No | 30000 | Request timeout in milliseconds |

Examples

Store a memory:

The agent calls store_memory with content text. The A-Mem engine auto-generates keywords, tags, and context, then asynchronously discovers links to related memories.

Search with filters:

Register custom dimensions first with register_dimensions (e.g., ["session_id", "project_id"]), then pass them as filters in search_memories to scope results.

Ingest a document:

Call ingest_document with the full text. The system chunks it and stores each chunk as a separate memory. Use get_job_status to track progress.

Development

npm install
npm run build        # Compile TypeScript
npm run dev          # Run with tsx (hot reload)
npm run inspector    # Open MCP Inspector UI

License

MIT