How to let Claude Code drive a browser with agent-browser

agent-browser is a native Rust CLI from Vercel Labs. It reads a page as an accessibility-tree snapshot and tags each element with a ref such as @e1, so an agent can click and type one command at a time. The browser stays open between commands, so each step is a single shell call.

1. Install it and download Chrome for Testing (first time only).

npm install -g agent-browser
agent-browser install

1. Add the skill so Claude Code knows how to use it. The installed skill is a thin stub that loads agent-browser skills get core, so it always matches the CLI version.

npx skills add vercel-labs/agent-browser

1. Try it by hand once.

agent-browser open example.com
agent-browser snapshot -i
agent-browser click @e1
agent-browser get url
agent-browser close

Refs change when the page changes, so take a new snapshot before reusing them. A click fails, rather than reporting success, when the target is covered by something like a consent banner.