mcp-graphiti

Graphiti MCP Server is a multi-graph platform that spins up a root server alongside project-specific MCP servers, all sharing a single Neo4j backend. The root server provides a central API surface and project orchestration, while each project container hosts its own graph extraction rules, entities, and models. Once running, you can access status at the root, and connect project tools to the per-project endpoints. The system includes editor auto-discovery, crash containment, and hot-reload capabilities so you can iterate on configurations without taking the whole stack down.

Prerequisites:

• Docker and Docker Compose installed on your machine
• Python (for initial setup and CLI tooling) or use the provided Docker images if you prefer a fully containerized workflow
• An OpenAI API key for text/entity extraction
• Neo4j instance accessible with credentials defined in .env

Installation steps:

1. Install the CLI and clone the repo

pipx install 'git+https://github.com/rawr-ai/mcp-graphiti.git'
# or clone manually
# git clone https://github.com/rawr-ai/mcp-graphiti.git

2. Prepare environment

cp .env.example .env  # edit with your Neo4j credentials and OpenAI key

3. Build and start containers

# If using the provided binaries/images, ensure Docker is running
# Start the root server and project servers (in development, you can run broker configs via the CLI)
graphiti compose   # generates docker-compose.yml and updates .cursor/mcp.json
graphiti up -d

4. Initialize a project (example)

cd /path/to/my-kg
graphiti init my-kg        # writes ai/graph/mcp-config.yaml
# add entity definitions under ai/graph/entities/

5. Rebuild/reload when needed

graphiti compose && graphiti up -d
graphiti reload <container-name>

Notes and tips:

• The root server runs on port 8000; project containers are exposed on 8001 and beyond as configured by your docker-compose setup.
• Ensure .env contains a strong NEO4J_PASSWORD; the server will not start with a weak password unless GRAPHITI_ENV=dev is set (not recommended for production).
• To keep projects isolated, each project gets its own group_id and model definitions while sharing the same Neo4j instance.
• If you need to wipe all projects, be cautious with NEO4J_DESTROY_ENTIRE_GRAPH=true as it removes every project on the next up cycle.
• Hot reload allows you to tweak a project's config and run graphiti reload <container> without downtime for other projects.