Skip to content
Get packs

Guide · Agent skill

How to use the Cognee Integrations Setup agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The cognee-integrations skill is an agent skill for the cognee memory system from topoteretes/cognee. Once installed, it teaches your AI agent how to connect cognee to other services: switching the language model or embedding provider, changing databases, using S3 storage, and running the MCP server for IDE integration. This guide explains what an agent skill is, how to install this one, how to trigger it, and what to check before you change settings.

Cognee Integrations Setuptopoteretes/cognee · Apache-2.0

Quick start

  1. 1Download and unzip the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. Alternatively, upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
  2. 2Prepare your environment. Install cognee with the extra for your backend, and have your provider API key and any Docker or AWS credentials ready.
  3. 3Ask for a specific change. Describe the integration you want, such as a provider, database or the MCP server, in plain language.
  4. 4Check .env.template. Ask the agent to verify variable names against the repo's .env.template before editing your .env.
  5. 5Configure LLM and embeddings together. Set the LLM and embedding provider variables so neither silently falls back to OpenAI.
  6. 6Reset if you changed embeddings. If you switched the embedding provider or model, reset local state and re-ingest your data.
  7. 7Test the connection. Run a small ingest and query to confirm the new database, storage or MCP connection works.

What is an agent skill, and what does this one do?

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent (such as Claude) can load when a task matches the skill's description. This skill covers connecting cognee to external services. It is not a program you run yourself: it gives the agent guidance so it can set up the right configuration for you.

  • Switch LLM providers: OpenAI, Azure, Gemini, Anthropic, Ollama, OpenRouter, custom/vLLM, or AWS Bedrock.
  • Configure embedding providers separately from the LLM, since the two are set independently.
  • Change databases: Postgres and pgvector, Neo4j, Neptune, Turso, or the default SQLite and LanceDB/LadyBug options.
  • Set up S3 storage and point data directories at s3:// paths.
  • Run the cognee MCP server so IDEs such as Cursor, Claude Desktop or Claude Code can use cognee memory.

How do you install the cognee-integrations skill?

Download the skill ZIP from this page and unzip it. You can install it for yourself or for a single project:

Personal install: unzip into ~/.claude/skills/ so the skill is available across your projects.

Project install: unzip into the .claude/skills/ folder inside your project, so it applies only there.

In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

How do you trigger the skill with natural requests?

You do not need a special command. Describe what you want in plain language and mention the service involved, because the skill's description matches requests about switching LLM or embedding providers, changing databases, S3 storage, or the MCP server. The skill is most useful when your request names a concrete target.

  • "Switch cognee to use Anthropic for the LLM and keep OpenAI for embeddings."
  • "Set up cognee with Postgres and pgvector for this project."
  • "Connect cognee's MCP server to Cursor."
  • "Configure cognee to store data in S3."

What do you need before you start?

The skill assumes a working Python environment and the cognee package. Gather these before asking the agent to make changes:

  • Python with the cognee package, plus the matching extra for your backend (for example cognee[postgres] or cognee[anthropic]).
  • An API key for your chosen LLM or embedding provider, such as LLM_API_KEY.
  • Docker and docker-compose if you use the Postgres, Neo4j or MCP profiles.
  • AWS credentials if you use S3 storage or Bedrock.
  • Redis or another service only if you choose it as the cache backend.
pip install cognee[postgres]

What does the skill check and warn about?

The skill points the agent to .env.template at the repo root as the authoritative list of variables, so the agent should check it rather than invent names. It also warns about common mistakes. The most important one: LLM and embedding providers are configured independently, so changing only the LLM leaves embeddings on OpenAI. Either keep a valid OpenAI key or configure both.

After changing the embedding provider or model, existing embeddings are no longer comparable. The skill advises resetting local state and re-ingesting. It describes two ways to do this: a full reset with cognee-cli forget --all, or dropping only the graph and vectors with cognee.forget(dataset=..., memory_only=True) while keeping your ingested files.

  • Configure both LLM and embedding providers when you switch either one.
  • Install the matching extra before switching a backend.
  • Reset and re-ingest after changing the embedding model.

What about safety: scripts and network access?

This skill does not run scripts, but it does use the network, because the configured providers and services are remote. It handles secrets: integration settings live in .env files and include API keys and cloud credentials. Do not paste real keys or credentials into a shared or public chat. Keep them in a local .env file that is not committed to version control.

The skill itself lists some variable names that may change, so confirm them against .env.template before you rely on them.

Example: running the cognee MCP server for an IDE

The skill describes starting the MCP server with the repo's Docker Compose profile. The server runs on port 8001 using Streamable HTTP at the /mcp path. Point your IDE's MCP settings at that URL. Configure the server's DB_* environment variables to match the main cognee service so both see the same data.

docker compose --profile mcp up

What people use it for

  • Setting up a new project to use Anthropic or a local Ollama model instead of the default OpenAI.
  • Deploying cognee with Postgres and pgvector and matching environment variables for a team.
  • Connecting cognee's MCP server to Cursor or Claude Code so the IDE assistant can use cognee memory.
  • Moving raw data storage to S3 for a shared deployment.

Frequently asked questions

Is the cognee-integrations skill free to use?

The skill is licensed under Apache-2.0, as part of the topoteretes/cognee repository. The providers and cloud services it configures, such as OpenAI, AWS or Postgres hosting, may charge for their own use, and this page does not give their prices.

What do I need to use the cognee-integrations skill?

You need Python with the cognee package and the matching extra for your backend, plus an API key for your chosen LLM or embedding provider. Docker and docker-compose are needed for the Postgres, Neo4j or MCP profiles, and AWS credentials are needed for S3 or Bedrock.

Where do I install an agent skill like this one?

Unzip the downloaded ZIP into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder for one project. In the Claude apps you can upload the ZIP under Settings → Capabilities → Skills.

Why does changing my LLM not change my embeddings?

LLM and embedding providers are configured separately, using LLM_* and EMBEDDING_* variables. If you set only the LLM, embeddings stay on OpenAI, so you must either keep a valid OpenAI key or configure the embedding provider too.

Do I have to delete my data after switching embedding models?

The source says embeddings from different models are not comparable, so after changing the embedding provider or model you should reset local state and re-ingest. To keep your ingested files, you can drop only the graph and vectors with cognee.forget(dataset=..., memory_only=True).

Does this skill work with the default OpenAI and SQLite setup?

Not really. The default setup needs only LLM_API_KEY, so the skill adds little for that case. It is most useful when you want a non-default model, database, storage or MCP server.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you are setting up cognee for a new project with a non-default model, a Postgres or Neo4j database, S3 storage, or the MCP server for an IDE. Skip it if you only need the default OpenAI and SQLite/LanceDB setup, and be careful with API keys and cloud credentials, which should never go into a shared chat.

Download the skill