Browser Use

Playwright browser automation via persistent CLI daemon. Fine-grained control with ~50ms latency per command.

Artifacts

Type	Detail
`tool`	`pip install browser-use && browser-use setup`
`skill`	browser-use/browser-use — CLI commands, workflow patterns

Requirements

Python 3.11+
Chrome/Chromium installed

What it does

CLI tool with a background daemon that keeps the browser alive between commands. Claude controls the browser directly via shell commands — navigate, click, type, extract data, take screenshots.

Key commands:

browser-use open <url> — navigate
browser-use state — discover clickable elements with indices
browser-use click <index> / browser-use input <index> "text" — interact by index
browser-use screenshot — capture page
browser-use eval "js code" — run JavaScript
browser-use get html --selector "css" — extract content

Browser modes: headless (default), --headed (visible), --profile (real Chrome with logins), cloud.

Setup

Ensure Python 3.11+ and Chrome are installed
Run harness-kit add browser-use
Run browser-use doctor to verify installation

vs agent-browser

	browser-use	agent-browser
Runtime	Python, Playwright	Node.js, Playwright
Approach	CLI daemon, element indices	Accessibility tree snapshots
Page representation	`state` command (clickable elements)	A11y tree (~200-400 tokens)
Screenshot	Built-in `screenshot` command	Not needed (text-based)
Session	Persistent daemon across commands	Per-command
Cloud option	`browser-use cloud connect`	No

Notes

Daemon keeps browser alive — no startup cost after first open
browser-use state is the key command — always run it first to get element indices
Aliases: bu, browser, browseruse
Skill source: skills.sh/browser-use/browser-use