Graphiti MCP Server • Fast Multi‑Project Knowledge Graphs

What is Graphiti MCP Server?

Graphiti MCP Server is a developer-focused fork of getzep/graphiti that extracts entities and relationships from text and stores them in Neo4j. It adds a CLI that spins up a root server plus project-specific MCP servers in Docker, enabling several knowledge graphs to share the same database with project isolation.

How to use Graphiti MCP Server?

Install with pipx, clone the repo, and fill in Neo4j credentials and OpenAI key in .env. Run graphiti compose to generate Docker config, then graphiti up -d to launch. Create projects with graphiti init <name> and add entity definitions. Connect MCP-compatible tools to the SSE endpoint at http://localhost:800{N}/sse.

Key features of Graphiti MCP Server

• Multi-project knowledge graphs sharing one Neo4j database
• Project isolation with different extraction rules per project
• Editor auto-discovery via .cursor/mcp.json port writing
• Crash containment — a bad prompt restarts only its container
• Hot reload for individual project configs
• Root server status endpoint at /graphiti/status

Use cases of Graphiti MCP Server

• Running multiple isolated knowledge graph projects on a single Neo4j instance
• Developing and testing extraction rules without cross-project interference
• Quickly iterating on entity definitions with hot reload
• Integrating with any MCP-compatible editor or tool via SSE transport

FAQ from Graphiti MCP Server

How is this different from the upstream Graphiti?

The upstream assumes one server per compose file. This fork manages many project servers in a single compose file that share Neo4j, adding project isolation, crash containment, and hot reload.

What are the runtime requirements?

A Neo4j instance, an OpenAI API key, and Docker. Python 3.10+ and pipx are needed for installation.

Where does my data live?

All extracted entities and relationships are stored in the Neo4j database configured in .env. No data is stored outside your own infrastructure.

What transport does the server use?

Projects expose Server-Sent Events (SSE) endpoints on ports starting at 8000. The root server runs on port 8000 by default.

How does authentication and security work?

The server refuses to start if NEO4J_PASSWORD is still set to password unless GRAPHITI_ENV=dev is enabled. Setting NEO4J_DESTROY_ENTIRE_GRAPH=true wipes all projects on the next graphiti up run.