Skip to main content
Your coding agent drives Boat with the boat CLI, the same as you. Give it three things:
  1. The CLI, signed in.
  2. The Boat docs, through the MCP server or llms.txt.
  3. A short rule in AGENTS.md or CLAUDE.md that says when to use a sandbox.
This page is about an agent on your computer. For the agents inside a sandbox, read Integrated agents.

1. Install the CLI and sign in

On Windows, run irm https://boat.dev/install.ps1 | iex in PowerShell. Open a new terminal after the install. An agent runs commands without a browser. Sign it in with an API key:
No account yet? The agent can make one itself. Read Sign up from an agent.

2. Give the agent the docs

Add the MCP server:
For Cursor, put the JSON in ~/.cursor/mcp.json, or in .cursor/mcp.json in your project. The ⋯ menu on every docs page also has one-click entries for Claude, ChatGPT, Cursor and VS Code.

3. Add a rule to AGENTS.md

Put this in AGENTS.md (Codex, Cursor and most agents) or CLAUDE.md (Claude Code) in your project:
AGENTS.md
The --json output of boat new is one JSON object per line. The ready line holds the id.

Skills

The open skill pack installs Boat skills for Claude Code, Codex, Cursor and other agents:
It has 13 skills, for example cloud-sandbox, parallel-agents, background-worker and public-url-hosting. Each skill tells the agent when to use Boat and which commands to run.
Some skills in the pack still show older values: the on.ascii.dev domain, an 80 GB disk and a fixed cap of 100 sandboxes. These docs are correct. Public URLs use on.boat.dev. For sizes and limits, read Pricing.
Inside a sandbox, every integrated agent already has a boat skill. You do not install it.