Graphiti

You run Graphiti’s MCP Server to expose its knowledge-graph capabilities to MCP-compatible clients. It supports streamable HTTP and legacy SSE transports, secure public deployments, queue monitoring, Redis-backed queues, and a rich set of tools to manage episodes, entities, and graph maintenance. This guide shows practical steps to use, install, and operate the server in common scenarios.

How to use

Connect from your MCP-compatible client using the Streamable HTTP transport at http://localhost:8000/mcp or the Legacy SSE transport at http://localhost:8000/sse. Use a group_id to namespace or isolate data when needed. You can perform operations such as adding episodes, searching nodes, and clearing or inspecting the graph. Use X-Group-Id headers to control multi-tenant access when using HTTP transports.

How to install

Prerequisites you need before starting: Python 3.10 or higher, a Neo4j database (Neo4j 5.26+ recommended), and an OpenAI API key for LLM operations.

Step-by-step setup and run flow:

# Prerequisites check
python --version
neo4j help

Clone the MCP server repository and navigate into the project folder.

git clone https://github.com/michabbb/graphiti-mcp-but-working.git
cd graphiti-mcp-but-working

Set up environment variables and dependencies. You can start from a local environment with uv as the dependency manager for reproducible builds.

# Install uv if you don't have it
curl -LsSf https://astral.sh/uv/install.sh | sh

# Create a virtual environment and install dependencies
uv sync

If you prefer Docker Compose, configure a .env file and use docker compose to start the services.

# Copy example env and customize values
cp .env.example .env
# Edit .env to set OPENAI_API_KEY, MODEL_NAME, and other options

# Start services
docker compose up

Additional setup for direct server run (optional)

You can run the MCP server directly with uv for a quick development workflow.

uv run python -m graphiti_mcp_server

Available tools

add_episode

Add an episode to the knowledge graph in text, JSON, or message formats.

search_nodes

Search the knowledge graph for relevant node summaries.

search_facts

Search for relevant facts (edges) between entities.

delete_entity_edge

Delete a specific entity edge from the graph.

delete_episode

Delete a specific episode from the graph.

delete_everything_by_group_id

Atomically delete all data (episodes, nodes, and edges) for a group_id.

get_entity_edge

Retrieve an entity edge by its UUID.

get_episodes

Fetch the most recent episodes for a given group.

get_queue_status

Show the status of episode processing queues and per-group details.

clear_graph

Clear the graph and rebuild indices. Requires CLEAR_GRAPH_PASSWORD.

get_status

Get the status of the MCP server and database connection.

list_group_ids

Discover and list all group_ids across nodes and relationships.

delete_everything_by_group_id

Atomic deletion of all data for a specific group_id.

delete_everything_by_group_id

Atomic removal of episodes, nodes, and edges for a group_id.

get_queue_status

Monitor background processing queues after memory insertion.