core-memory-mcp

# ✨ CoreMemory-MCP A local, MCP-compliant, long-term memory *service* for AI agents. --- **[Read this in Chinese (中文文档)](README_zh.md)** --- ## 💡 What is CoreMemory-MCP? CoreMemory-MCP is a background service that provides a simple and persistent long-term memory for any AI agent that supports the **Model Context Protocol (MCP)**. It allows tools like **Cursor**, **Gemini-CLI**, and **Claude Code-Cli** to retain and recall information across sessions. It runs as a standalone server on your local machine, acting as a language-agnostic, pluggable brain for your favorite AI tools. ## ⚙️ How It Works The architecture is a standard client-server model based on the MCP standard: `[Your Agent / IDE] <--> [MCP (JSON-RPC over HTTP)] <--> [Core Memory Service]` 1. You run the Core Memory service. 2. You configure your client application (e.g., Cursor, Gemini-CLI) to connect to the service. 3. Your client application can then discover and execute the memory tools (`memory.add`, `memory.search`, etc.) provided by the service. ## 🚀 Installation > ⚠️ **Note**: The package is not yet published to PyPI. The following are the planned installation methods. It is recommended to install the package via pip once it is published: ```bash pip install core-memory-mcp ``` Alternatively, for development, you can clone the repository and install it in editable mode: ```bash git clone https://github.com/michaelfeng/CoreMemory-MCP.git cd CoreMemory-MCP pip install -e . ``` ## ▶️ Running the Service If you installed the package via pip, you can start the service by running the following command in your terminal: ```bash core-memory-server ``` If you are running from the source code, you can also run the script directly: ```bash python memory_service.py ``` You will see output indicating the service is running and ready to accept connections: ``` --- Starting Core Memory MCP Service --- Compatible with the Model Context Protocol. Listening on http://127.0.0.1:5001 Acces…