Graphiti MCP Server

# Graphiti MCP Server - Project Structure This document describes the structure and key components of the standalone Graphiti MCP Server project. ## Project Overview The Graphiti MCP Server is a standalone Model Context Protocol (MCP) server that exposes Graphiti's knowledge graph capabilities through standardized MCP tools and resources. It can be used with any MCP-compatible AI assistant or IDE. ## Directory Structure ``` graphiti-mcp-standalone/ ├── graphiti_mcp_server.py # Main server implementation ├── pyproject.toml # Project metadata and dependencies ├── README.md # Project documentation and usage guide ├── LICENSE # Apache 2.0 License ├── Dockerfile # Docker image definition ├── docker-compose.yml # Docker Compose configuration (includes Neo4j) ├── .env.example # Environment variable examples ├── example_usage.py # Example usage demonstration ├── cursor_rules.md # Cursor IDE integration rules ├── mcp_config_sse_example.json # SSE transport configuration example ├── mcp_config_stdio_example.json # STDIO transport configuration example ├── TESTING_GUIDE.md # Testing guide and instructions └── uv.lock # Dependency lock file ``` ## Key Components ### Main Server (`graphiti_mcp_server.py`) The core implementation that provides: - **MCP Tools**: Exposes Graphiti functionality as MCP tools - `add_memory`: Add episodes to the knowledge graph - `search_memory_nodes`: Search for entity nodes - `search_memory_facts`: Search for relationships/facts - `delete_entity_edge`: Delete entity relationships - `delete_episode`: Delete episodes - `get_entity_edge`: Retrieve specific entity relationships - `get_episodes`: Get recent episodes - `clear_graph`: Clear the entire knowledge graph - **MCP Resources**: Provides server status information - `http://graphiti/status`: Server and database connection status - **Transports**: Supports multiple communication protocols - STDIO: For local AI assistants (Claude Desktop, etc.) - HTTP: For web-based integrations - SSE: For streaming connections (Cursor, etc.) ### Configuration Files - **pyproject.toml**: Defines project metadata, dependencies, and entry points - **Dockerfile**: Container image definition for easy deployment - **docker-compose.yml**: Complete deployment with Neo4j database - **.env.example**: Environment variable templates ### Documentation - **README.md**: Comprehensive usage guide and integration examples - **cursor_rules.md**: Specific integration instructions for Cursor IDE - **TESTING_GUIDE.md**: Testing procedures and guidelines ## Installation and Usage ### Quick Start ```bash # Install the package pip install graphiti-mcp-server # Or install from source pip install -e . # Run the server graphiti-mcp-server --transport sse --group-id my_project ``` ### Docker Deployment ```bash # Build and run with Docker Compose (includes Neo4j) docker-compose up ``` ## Integration Examples The server can be integrated with various MCP-compatible clients: - **Cursor IDE**: Via SSE transport - **Claude Desktop**: Via STDIO transport with mcp-remote gateway - **Custom AI Applications**: Via any MCP transport ## Dependencies Key dependencies include: - `graphiti-core`: The core knowledge graph library - `fastmcp`: MCP protocol implementation - `openai`: LLM client library - `neo4j`: Graph database driver - `azure-identity`: Azure authentication support - `python-dotenv`: Environment variable management ## Development The standalone server is designed to be: - **Self-contained**: All necessary components included - **Configurable**: Flexible environment variable system - **Deployable**: Multiple deployment options (pip, Docker, source) - **Compatible**: Works with standard MCP clients - **Extensible**: Easy to add new tools and features