security. ENABLE_BACKEND_ACCESS_CONTROL=false is the single-user local posture; configure authentication before going beyond a local demo.
Who it is forA team lead who wants one shared memory service for several agents.
The problemEach agent keeping its own memory leads to duplicated and conflicting context.
The outcomeCognee running with the API on 8000, the UI on 3000 and MCP on 8001.
What you need first
- Python 3.10 to 3.14; install with `uv pip install "cognee[gliner]"` for the local extraction model
- Optional: an LLM key (LLM_API_KEY) for generated answers instead of retrieved passages
- Optional: Docker for the API, UI and MCP server; Node.js/npm for the local UI launcher
Steps
- Try the bundled graph first with
cognee-cli demo - Run the API image with
docker run --rm -it -p 8000:8000and the LLM_API_KEY, ENABLE_BACKEND_ACCESS_CONTROL and volume flags shown in the README - From a source checkout, copy .env.template to .env and run
docker compose --profile ui --profile mcp up - Open the local UI with
cognee-cli -uito inspect the graph
Prompt for your agent
Using topoteretes/cognee, start the stack with `docker compose --profile ui --profile mcp up` from a checkout, load <my docs>, and confirm the API on port 8000 and MCP on port 8001 respond.
Caveat. Running the whole memory layer on Postgres is a demo feature per the README.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for topoteretes/cognee and "Self-host the Cognee API, UI and MCP server"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer