MCP Mem

🚀 MCP Memory

An MCP (Model Context Protocol) server that implements a memory solution for data-rich applications and manages memory through efficient graph knowledge graph capabilities.

🚀 Quick Start

This MCP server implements a memory solution for data-rich applications that involve obtaining information from multiple sources, including uploaded files. It uses HippoRAG internally to manage memory, which is a required dependency for this package.

✨ Features

• Session-based Memory: Create and manage memory for specific chat sessions.
• Efficient Knowledge Graph: Use HippoRAG for memory management.
• Multiple Transport Support: Compatible with stdio and SSE transports.
• Search Functionality: Search for information from various sources, including uploaded files.
• Automatic Resource Management: TTL-based cleanup for sessions and memory instances.

📦 Installation

Install from PyPI:

pip install mcp-mem hipporag

Or install from source:

git clone https://github.com/ddkang1/mcp-mem.git
cd mcp-mem
pip install -e .
pip install hipporag

⚠️ Important Note

HippoRAG is a required dependency for mcp-mem to function properly.

💻 Usage Examples

Basic Usage

You can run the MCP server directly:

mcp-mem

By default, it uses the stdio transport. To use the SSE transport:

mcp-mem --sse

You can also specify the host and port for the SSE transport:

mcp-mem --sse --host 127.0.0.1 --port 3001

📚 Documentation

Basic Configuration

To use this tool in combination with Claude in Windsurf, add the following configuration to the MCP configuration file:

"memory": {
    "command": "/path/to/mcp-mem",
    "args": [],
    "type": "stdio",
    "pollingInterval": 30000,
    "startupTimeout": 30000,
    "restartOnFailure": true
}

The command field should point to the location where you installed the Python package using pip.

Environment Variable Configuration

You can configure the LLM and embedding models used by mcp-mem through environment variables:

• EMBEDDING_MODEL_NAME: The name of the embedding model to use (default: "text-embedding-3-large").
• EMBEDDING_BASE_URL: The base URL of the embedding API (optional).
• LLM_NAME: The name of the LLM model to use (default: "gpt-4o-mini").
• LLM_BASE_URL: The base URL of the LLM API (optional).
• OPENAI_API_KEY: The OpenAI API key (required).

Memory Management Configuration

• Session TTL: Set to 0 to disable cleanup. The default value is 0.
• Instance TTL: Set to 0 to disable cleanup. The default value is 0.

Provided Tools

• Create Memory: Use create_memory to create new memory.
• Store Memory: Use store_memory to store memory.
• Retrieve Memory: Use retrieve_memory to retrieve memory.

🔧 Technical Details

To develop, install the following dependencies:

pip install -r requirements.txt

Then run the tests:

pytest tests/

Refer to the following for code style:

black .
flake8 .
mypy .

🤝 Contribution

Contributions should follow the standard development process, and all tests should pass before submission.

📄 License

This project is licensed under the MIT License.