C

Cognee

Turn your data into a knowledge graph that gives agents long-term memory.

Works with: Claude DesktopClaude CodeCursorWindsurf

How to install the Cognee MCP server

Add this to your Claude Desktop MCP configuration:

Run from source: clone github.com/topoteretes/cognee and start the cognee-mcp server (Docker image available). Requires an LLM API key.

Add this to your Claude Code MCP configuration:

Run from source: clone github.com/topoteretes/cognee and start the cognee-mcp server (Docker image available). Requires an LLM API key.

Add this to your Cursor MCP configuration:

Run from source: clone github.com/topoteretes/cognee and start the cognee-mcp server (Docker image available). Requires an LLM API key.

Add this to your Windsurf MCP configuration:

Run from source: clone github.com/topoteretes/cognee and start the cognee-mcp server (Docker image available). Requires an LLM API key.

Built by ContextBoltCognee gives your agent one kind of memory. ContextBolt Bookmarks adds another: everything you save across X, Reddit, and LinkedIn, searchable by meaning.

See ContextBolt Bookmarks

Cognee, from topoteretes, is an official MCP server that turns your data into a knowledge graph so agents have persistent long-term memory. If you are building an agent that needs to remember and reason across information over time, this gives it a structured place to store and recall it.

Why use it

Most agents forget everything between sessions. Cognee ingests your data, builds a knowledge graph from it, and lets the agent query that graph later, so context survives past a single conversation and relationships between facts stay intact.

What it actually does

You feed it data, it structures that into a graph, and the agent reads from it as memory. Example prompts:

  • ‘Ingest these meeting notes into memory and link the people mentioned.’
  • ‘What do we know about this customer from everything ingested so far?’
  • ‘Recall the decisions we made about the pricing model.‘

Gotchas

This is a developer tool, not a one-click install. You run it from source, and it needs an LLM API key to process and structure data, which means ongoing model costs. A Docker image is available to simplify setup. Building good memory still takes thought about what you ingest and how you query it, so expect setup and tuning work rather than instant results.

Built by ContextBolt

Combine Cognee with ContextBolt Bookmarks

Cognee gives Claude one kind of memory. ContextBolt Bookmarks adds another: every tweet, post, and article you save across X, Reddit, and LinkedIn becomes searchable by meaning. Run both as MCP servers and Claude can pull from both layers in one prompt.

See ContextBolt Bookmarks →

Cognee MCP server: FAQs

How do I run it?

You run it from source, with a Docker image available to make setup easier. It is aimed at developers rather than non-technical users.

Does it need an API key?

Yes. It requires an LLM API key to process and structure the data it ingests, which adds ongoing model costs.

What does it actually store?

It builds a knowledge graph from the data you ingest, capturing entities and the relationships between them for later recall.

Who makes it?

It is the official Cognee server from topoteretes, the team behind the project.

Is it plug-and-play?

No. Expect setup and tuning. Useful memory depends on what you ingest and how you structure your queries.