MemGPT MCP Server

What is MemGPT MCP Server?

A TypeScript-based MCP server that implements a memory system for LLMs. It provides tools for chatting with different LLM providers (OpenAI, Anthropic, OpenRouter, Ollama) while maintaining conversation history.

How to use MemGPT MCP Server?

Install dependencies with npm install, build with npm run build, and configure in Claude Desktop by adding the server config with required environment variables (OPENAI_API_KEY, ANTHROPIC_API_KEY, OPENROUTER_API_KEY). Use the provided tools to chat, retrieve memory, clear memory, switch providers, and switch models.

Key features of MemGPT MCP Server

• Send messages to the current LLM provider via chat tool.
• Retrieve conversation history with optional limit via get_memory.
• Clear all stored memories via clear_memory.
• Switch between OpenAI, Anthropic, OpenRouter, and Ollama via use_provider.
• Switch provider-specific models (e.g., Claude 3/3.5 series, GPT-4, Ollama models) via use_model.
• Provider and model selections persist across sessions.

Use cases of MemGPT MCP Server

• Chat with multiple LLM providers while keeping a persistent conversation history.
• Retrieve and review past conversations (full or limited) for context continuity.
• Clear conversation memory to reset or start a new session.
• Dynamically switch providers or models without restarting.

FAQ from MemGPT MCP Server

What LLM providers are supported?

OpenAI, Anthropic, OpenRouter, and Ollama.

How do I retrieve the full conversation history?

Use the get_memory tool with { "limit": null } to retrieve all stored memories. Default limit is 10 if not specified.

What models are available for Anthropic?

Anthropic Claude 3 series (Haiku, Sonnet, Opus) and Claude 3.5 series (Haiku, Sonnet). Use model names like claude-3-haiku or claude-3.5-sonnet.

Where is configuration stored for Claude Desktop?

On MacOS at ~/Library/Application Support/Claude/claude_desktop_config.json, on Windows at %APPDATA%/Claude/claude_desktop_config.json.

How can I debug the server?

Use the MCP Inspector by running npm run inspector, which provides a URL for browser-based debugging tools.