Memory server for AI Chat with MCP interface
@Gelembjuk
This is the MCP server with memory interface. It can be used with an AI Chat tool as a memory service
Overview
What is Memory server for AI Chat with MCP interface?
It is an example memory server for AI chat that follows the MCP (Model Context Protocol) interface supported by the tool CleverChatty. It exposes two tools — remember (store a chat message with a role) and recall (return a summary of previous conversations) — and is designed to be used alongside the CleverChatty CLI for chat applications with persistent memory.
How to use Memory server for AI Chat with MCP interface?
Clone the repository, install uv, create a virtual environment, run uv sync, then start the server with fastapi run mcp_server.py --port 8001. The server uses SSE transport and is accessible at http://localhost:8001/mcp. A CLI (python manager.py) is provided for testing and debugging, supporting commands like remember, recall, clear-memory, patch-memories, and history-dump.
Key features of Memory server for AI Chat with MCP interface
- Exposes
rememberandrecallMCP tools. - Stores chat messages with role and content.
- Returns a summary of previous conversations.
- Runs as a FastAPI server with SSE transport.
- Includes a CLI for manual testing and debugging.
Use cases of Memory server for AI Chat with MCP interface
- Provide memory to an AI chat application using the CleverChatty interface.
- Store and retrieve conversation context for multi-turn dialogues.
- Test and debug memory behavior via the included CLI commands.
FAQ from Memory server for AI Chat with MCP interface
How do I start the server?
Clone the repo, install uv, create a virtual environment, run uv sync, then execute fastapi run mcp_server.py --port 8001. The server will listen on port 8001 (default localhost only).
What tools does the server expose?
Two tools: remember (requires role and message strings) and recall (no input parameters). remember stores a message, recall returns a summary of previous conversations.
How can I test the server without an MCP client?
Use the provided CLI: python manager.py COMMAND, for example python manager.py remember user "Hello" or python manager.py recall.
Can the server be accessed from other machines?
Yes. Start the server with --host 0.0.0.0 to make it accessible from any IP address. By default it only listens on localhost.
What transport protocol does the server use?
It uses SSE (Server-Sent Events) transport. The MCP endpoint is http://localhost:8001/mcp (or the configured host/port).
