MCP-Mem0: Long-Term Memory for AI Agents

What is MCP-Mem0?

MCP-Mem0 is a reference implementation of an MCP (Model Context Protocol) server integrated with the Mem0 memory platform, providing AI agents with persistent, searchable long-term memory. It is designed for developers building MCP-compatible agents that need to store, retrieve, and semantically search information across sessions.

How to use MCP-Mem0?

Clone the repository, install dependencies with uv or build the Docker image, configure environment variables (transport, LLM provider, database URL), and run the server via uv run src/main.py or docker run. Connect any MCP client by pointing it to the SSE endpoint or by configuring stdio transport in the client's MCP settings.

Key features of MCP-Mem0

• Semantic indexing of stored information for relevance-based retrieval
• Three memory management tools: save, list all, and search memories
• Supports SSE and stdio transport protocols
• Compatible with multiple LLM providers (OpenAI, OpenRouter, Ollama)
• Docker-based deployment for containerized environments

Use cases of MCP-Mem0

• AI assistants that need to remember user preferences and conversation history across sessions
• Agents performing multi-step research that require persistent context
• Knowledge management tools that organize and retrieve information via natural language queries

FAQ from MCP-Mem0

What dependencies does MCP-Mem0 require?

Python 3.12+, a PostgreSQL database (with Supabase supported) for vector storage, and API keys for your chosen LLM provider (OpenAI, OpenRouter, or Ollama). Docker is recommended for containerized deployments.

Which transports does MCP-Mem0 support?

SSE (Server-Sent Events) and stdio (standard input/output) transport protocols are both supported, configurable via the TRANSPORT environment variable.

What LLM providers and models can I use with MCP-Mem0?

OpenAI, OpenRouter, and Ollama are supported. The provider, base URL, API key, model, and embedding model are all configurable through environment variables.

How does MCP-Mem0 store and retrieve memories?

Memories are stored with semantic indexing using embedding vectors in a PostgreSQL database. The server offers three tools: save_memory to store information, get_all_memories to retrieve everything, and search_memories to find relevant memories via semantic search.

How do I connect an MCP client to MCP-Mem0?

For SSE transport, use the URL http://localhost:8050/sse (adjust host/port as needed). For stdio transport, configure the client to run the Python script or Docker container with the required environment variables.