READMEmain

Graphiti MCP Server

A knowledge graph memory system for AI agents using the Model Context Protocol (MCP).

Based on Graphiti, an open-source knowledge graph for AI agents.

Overview

Graphiti MCP Server provides a temporal memory graph for AI agents, allowing them to store and retrieve information with rich context. It uses:

Graphiti: Core graph functionality for managing entities and relationships
Neo4j: Graph database for storing knowledge
Graphiti: Core graph functionality for managing entities and relationships
MCP Server: Model Context Protocol interface for AI agents to interact with the graph

Prerequisites

Docker and Docker Compose
OpenAI API key

Setup Instructions

Clone the repository

git clone <repository-url>
cd GrafitiMcp

Set up environment variables

cp .env.template .env

Edit the .env file and add your OpenAI API key and other configuration:

NEO4J_PASSWORD=your_secure_password_here
OPENAI_API_KEY=your_openai_api_key_here
MODEL_NAME=gpt-4o  # or another OpenAI model

Start the services
```
docker-compose up
```
This will start:
- Neo4j database on port 8044
- Graphiti API on port 8080
- Graphiti MCP Server on port 8001

Connecting to the MCP Server

The MCP server runs on port 8001 and uses Server-Sent Events (SSE) transport by default.

In Roo Code

To connect to the MCP server in Roo Code:

Open Roo Code
Go to Settings > MCP Servers
Add a new MCP server with:
- Name: graphiti-mcp
- URL: http://localhost:8001
- Transport: sse

In Custom Applications

To connect to the MCP server in your own applications, use the MCP client library:

from mcp.client import MCPClient

# Create an MCP client
client = MCPClient("http://localhost:8001", transport="sse")

# Connect to the server
await client.connect()

# Use the tools provided by the server
result = await client.use_tool("add_episode", {
    "name": "Test Episode",
    "episode_body": "This is a test episode",
    "source": "text"
})

Basic Usage Examples

Adding Information to the Graph

# Add a text episode
await client.use_tool("add_episode", {
    "name": "Company News",
    "episode_body": "Acme Corp announced a new product line today.",
    "source": "text",
    "source_description": "news article"
})

# Add structured JSON data
await client.use_tool("add_episode", {
    "name": "Customer Profile",
    "episode_body": "{\"company\": {\"name\": \"Acme Technologies\"}, \"products\": [{\"id\": \"P001\", \"name\": \"CloudSync\"}, {\"id\": \"P002\", \"name\": \"DataMiner\"}]}",
    "source": "json",
    "source_description": "CRM data"
})

Searching the Graph

# Search for nodes
nodes = await client.use_tool("search_nodes", {
    "query": "What products does Acme Technologies offer?",
    "max_nodes": 5
})

# Search for facts
facts = await client.use_tool("search_facts", {
    "query": "Tell me about Acme Corp's product announcements",
    "max_facts": 5
})

Troubleshooting

Connection Issues

If you can't connect to the MCP server:

Ensure all services are running: docker-compose ps
Check the logs: docker-compose logs graphiti-mcp
Verify the server is accessible: curl http://localhost:8001/status

Neo4j Connection Errors

If the MCP server can't connect to Neo4j:

Check Neo4j is running: docker-compose logs neo4j
Verify Neo4j credentials in the .env file
Ensure Neo4j is healthy: docker-compose exec neo4j cypher-shell -u neo4j -p <your-password> "RETURN 1;"

License

[Specify your license here]

GrafitiMcp