Deploy with AI

Tell an AI coding agent — Claude Code, Claude.ai, Cursor, ChatGPT — "deploy this to Koo" and watch it handle the rest: create the service, deploy it, follow the deployment until it is ready, and report your live URL. It works through the hosted Koo MCP server, which your agent connects to by URL.

Koo does not run an AI model for you. The agent is your own coding tool, driving Koo through the same API the console uses.

How it works

Point your agent at Koo's MCP endpoint. There is nothing to install — in Claude Code it is one command:
text
```
claude mcp add --transport http koo https://mcp.koo.io/mcp
```
In Claude.ai, Cursor, or ChatGPT, add a connector with the same URL.
Authorize it. The first tool call opens your browser on Sign in with Koo, where you pick one account and grant Read or Write — the screen states exactly what the agent will be able to do before you approve. There is no token to mint, paste, or store.
Prompt the agent — "deploy this to Koo". It creates the service, starts a deployment, follows the deployment timeline until the service is ready, and reports the live URL.

To deploy from local source rather than a prebuilt image, the agent has you run koo up in your project directory (the CLI packs and builds it), then watches the result with the same tools.

Where there is no browser to authorize in — a CI runner, a remote shell — the same endpoint accepts a kc_… API token as its bearer instead; see Headless / CI.

From there the service is a normal Koo service: the deployment appears in the console, and logs, metrics, and rollback work as usual.

Same API, same guardrails

The agent drives the exact endpoints the console calls — there is no separate deploy path. Your account's roles, plan limits, and audit trail apply to every action the agent takes. Actions are attributed to you, together with the client that made them, so you can always see what the agent did — and you can revoke the connection at any time from the console's Connections page.