MCPOfficialCognee
Cognee MCP
Long-term agent memory built on a knowledge graph: remember, recall and forget.
The MCP server for Cognee exposes agent memory through three tools – remember, recall and forget – storing data in a knowledge graph, while code_search queries indexed code (symbols, dependencies and architecture). It runs locally over stdio, SSE or HTTP or connects to Cognee Cloud; the default setup needs an LLM key, such as OpenAI.
Installation
Manual install
$docker run -e TRANSPORT_MODE=http --env-file ./.env -p 8000:8000 --rm -it cognee/cognee-mcp:mainDetails
- Works in
- Claude Code, Claude, Cursor, Codex, Gemini CLI, VS Code, Windsurf, Zed
- Transport
- local
- License
- Apache-2.0
- Tasks
- Research, Code
- Rating
- Source
- github.com/topoteretes/cognee/tree/main/cognee-mcp (opens in a new tab)
Tags
- memory
- knowledge graph
- rag
- agents
A pack is your own set of skills and MCP servers. sciver loads it in your AI with one command.