DNA: updated from connect-an-ai-agent.md

Root ·

Updated from docs/dna/connect-an-ai-agent.md in the Knowen repository: How Knowen works is kept with the code.

Changes to the content

# Connect an AI agent

**Goal:** let Claude or another AI assistant read and write your graphs, or call Knowen from a script.

**Before you start:** an assistant acts with the access you give it. Use a separate token for each tool, so you can revoke one without the others.

1. Open **Settings › API tokens**.
2. **For an assistant that speaks MCP**, open your client under **Connect via MCP** and follow its steps. Claude Desktop connects as a custom connector by signing in, with no token; Claude Code and others use an API token.

   ![API tokens: create a token, how to send it, and Connect via MCP](/dna/media/connect-an-agent.webp?v=56119c84)
3. **For a script**, click **New API token**, copy it (it is shown once), and send it with each request: `Authorization: Bearer <token>`.
4. **For an agent with an account of its own**, open **Settings › AI agents** and click **New agent**. It gets its own @handle and works only in the private graphs you give it.

**What happens next.** The assistant can read what you can read and propose changes as you would; its proposals go through the same AI first pass and owners' decision as anyone's. Writing in a graph you collaborate on through the API needs its owner's grant.

**Undo.** Revoke a token on the same page; an agent can be deactivated in **Settings › AI agents**.

# Parents

* Agents and the API

Feedback00

No feedback yet. Say whether this change improves the node.

Sign in to add a new comment.