AGENT MEMORY / Zep
Graphiti MCP Server
Keeps agent memory as a temporal knowledge graph in FalkorDB or Neo4j, using an LLM to extract entities and facts from each added episode.
“This is an experimental Model Context Protocol (MCP) server implementation for Graphiti.”
01 / THE REASONING
Why this made the selection.
- Published in Zep's getzep/graphiti repository under mcp_server/; the README at this commit lists thirteen tools, including add_memory, search_memory_facts, get_episode_entities, delete_episode and clear_graph.
- The README names the supported LLM providers (OpenAI, Anthropic, Gemini, Groq, Azure OpenAI) and embedders (OpenAI, Voyage, Sentence Transformers, Gemini), so extraction can run on the provider the user chooses.
02 / THE REVIEW RECORD
What we actually inspected.
Source review has boundaries.
A clear record is more useful than a “safe” badge.
Material inspected
- mcp_server/README.md at 9d56c190926b382a54c5ea03e7680af1d05e1c9d
- mcp_server/pyproject.toml at the same commit
- LICENSE at the same commit (the README says the server uses the parent project's licence)
Our findings
- The README's defaults are HTTP transport at http://localhost:8000/mcp/, FalkorDB in a combined container with the server, OpenAI model gpt-5.5 and text-embedding-3-small.
- On FalkorDB each group_id is stored in its own graph; tools called without group_id use the configured default group.
- mcp_server/pyproject.toml names mcp-server 1.1.0, Python >=3.10,<4, with optional provider extras.
Not established by this review
- docker compose up
- Any LLM or embedding call
- Any graph read or write
The review applies to the material and revision named here. A newer upstream release can change its behavior.
03 / PUT IT TO WORK
Connect a server deliberately.
DOCUMENTED COMMAND
docker compose up Copying does not execute this command. It may retrieve a newer version than the reviewed source.
- Clone getzep/graphiti, copy .env.example to .env in graphiti/mcp_server and set at least one LLM provider key there; the README requires the .env file in that directory. Do not commit keys.
- Run the command above from graphiti/mcp_server; the README says it starts FalkorDB and the MCP server in one container. Point the client at http://localhost:8000/mcp/. It was not run.
Before you start
- Docker and Docker Compose
- An OpenAI API key or another supported LLM provider key
Documented configuration example
Use the client and authentication context described in the setup guidance. Configuration formats vary between hosts.
{
"mcpServers": {
"graphiti-memory": {
"transport": "http",
"url": "http://localhost:8000/mcp/"
}
}
} 04 / FOLLOW THE EVIDENCE
The source trail.
Our notes are separate from the original resource.
Check upstream before adopting a new version.
- Graphiti MCP server README ↗Checked
https://github.com/getzep/graphiti/blob/9d56c190926b382a54c5ea03e7680af1d05e1c9d/mcp_server/README.md
Supports: Experimental status, Tools, Defaults and providers, Telemetry, Client configuration
- Graphiti LICENSE ↗Checked
https://github.com/getzep/graphiti/blob/9d56c190926b382a54c5ea03e7680af1d05e1c9d/LICENSE
Supports: Apache License 2.0
https://github.com/getzep/graphiti/blob/9d56c190926b382a54c5ea03e7680af1d05e1c9d/mcp_server/pyproject.toml
Supports: Package mcp-server 1.1.0, Provider extras