Graphiti MCP Server
Knowledge graph memory from Zep that tracks facts and how they change over time.
Officialby ZepRuns on your computerChecked October 18, 2026
The Graphiti MCP server, from the team behind Zep, turns what Claude learns into a time-aware knowledge graph of entities, facts and episodes stored in FalkorDB or Neo4j. It uses an LLM provider such as OpenAI or Anthropic to extract entities, and its maintainers still label the MCP server experimental.
Who it is for
- Developers building agents with long-term memory
- Teams that need to know when a fact changed
- Researchers experimenting with knowledge graphs
Prompts to try
Once it is connected, ask in plain words. Claude decides when to use the server and asks before it changes anything important.
Add this meeting summary to memory under the group acme-project.What facts do we have about the Acme contract, and which of them changed recently?Search memory for everything related to our hiring plan.How to connect it
In Claude Code
claude mcp add --transport http graphiti-memory http://localhost:8000/mcp/
In the Claude desktop app
Add this to claude_desktop_config.json, then restart the app:
{
"mcpServers": {
"graphiti-memory": {
"command": "npx",
"args": [
"mcp-remote",
"http://localhost:8000/mcp/"
]
}
}
}Good to know
Start the server first with docker compose up inside graphiti/mcp_server (it bundles FalkorDB), with OPENAI_API_KEY or another provider key set.
Unlock the setup steps
Free. Leave your email once to open the setup steps for all 150 servers, and get the best new connectors each month.
How we checked it
Checked the mcp_server folder of the getzep/graphiti repo (Apache-2.0, last commit 2026-10-15, release v0.31.1) and its README setup steps for Docker, FalkorDB and Neo4j.