Sign in
Docs

Connect an agent

Give Claude Code, Cursor, Codex and other agents the same memory your team uses.

An agent connected to trellismem can recall from your team’s memory and store what it learns. It acts with an API key, and sees exactly what the owner of that key can see.

1. Create an API key

Open API keys in the app and create a key. Create one per agent or automation, so you can revoke a single key later without disturbing the rest.

API keys skip the second sign-in step on purpose. That keeps agents working when the people who own them turn on two-step sign-in.

2. Follow the guided setup

Open Integrations. Each supported tool has a card with the exact steps for it, and the API address and your key already filled in:

  • Claude Code
  • Codex
  • Claude Desktop
  • Cursor
  • VS Code
  • Gemini CLI
  • Cline
  • Hermes Agent
  • OpenClaw
  • n8n
  • Dify

3. Or configure MCP by hand

Any tool that speaks the Model Context Protocol (MCP) can use the memory server. It runs through uvx, so the only requirement is uv.

{
  "mcpServers": {
    "cognee": {
      "command": "uvx",
      "args": ["cognee-mcp"],
      "env": {
        "COGNEE_BASE_URL": "https://api.trellismem.com",
        "COGNEE_API_KEY": "your-api-key"
      }
    }
  }
}

COGNEE_BASE_URL is the address of the trellismem API, not of the app. Keep the server name and the two setting names exactly as shown.

The Integrations page pins the server to the version that matches the service. Copy the config from there when you can.

4. Test it

Ask the agent:

What do you know from cognee?

It should call the memory tool and answer from one of your datasets. If it answers from general knowledge instead, the server is not connected. Check the address and the key first.