Browser Use

Browser Use CLI

Give Claude Code a browser. Paste one prompt.

The agent installs the Browser Use CLI, registers the skill, and starts driving your own Chrome, with your logins. The same commands open a cloud stealth browser when you need more than one.

Copy the promptCLI docs
(Install or upgrade browser-use with  `uv tool install --python 3.12 --upgrade --force 'browser-use @ git+https://github.com/browser-use/browser-use.git'`,run `browser-use skill install`, and connect it to my browser.Follow https://github.com/browser-use/browser-use if setup or connection fails.)
into Claude Code, Codex or CursorYour Chrome · Cloud browsers
1 prompt
installs the CLI and the skill
2 browsers
your own Chrome, or a cloud browser
$0.02 / hour
when you move to a cloud stealth browser; local Chrome is free

What happens after you paste it

  1. 1

    The agent installs and connects

    uv installs the CLI, the skill registers itself with the agent, and Chrome asks once whether remote debugging is allowed.
  2. 2

    You ask for browser work

    Test the signup flow, read a dashboard, fill a form, export a table. The skill steers the agent to the accessibility tree for targeting and to page state for verification, so it does not guess from pixels.
  3. 3

    It scales when you need it

    One cloud browser per parallel sub-agent, with stealth, proxies and recordings, from the same commands. Local for logged-in work, cloud for volume. Hermes Agent uses it as the default browser mode.

The CLI sits between the cloud and the Python library.

Browser Use comes three ways: the fully hosted cloud, the open-source Python agent, and the CLI in between, which gives an agent you already have, Claude Code, Codex, Cursor, Hermes, OpenClaw, a browser of its own. The agent runs short Python snippets inside the CLI: open a tab, click, type, read the accessibility tree, take a screenshot, run JavaScript, download a file. A skill file tells it when to reach for the browser and how to verify what it did.
Three ways to use Browser Use: fully hosted cloud; your existing agent with the Browser Use CLI; or the open-source Browser Use agent as a Python library. The CLI and the library each connect to a local or cloud browser.
From the browser-use README: hosted cloud, the CLI for your existing agent, or the Python library.

The CLI in five commands

browser-use skill install
Registers the skill with Claude Code, Codex, Cursor and other agents that read skill files.
browser-use <<'PY' … PY
Runs Python against the browser: new_tab, click_at_xy, type_text, page_info, js, and the accessibility tree over CDP.
browser-use auth login
Signs in to Browser Use Cloud so start_remote_daemon opens stealth browsers, one per parallel sub-agent.
browser-use --doctor
Diagnoses the Chrome connection when the remote-debugging permission or the daemon is off.
uvx browser-use
Runs the CLI once without installing it, for a quick try or a CI job.

Questions people ask

What does the agent get?

A CLI it runs Python inside: open tabs, click, type, read the accessibility tree, run JavaScript, take screenshots, handle downloads and uploads. A skill file tells the agent when to reach for it, so there are no tool definitions to write.

Does it use my own Chrome?

By default, yes. It attaches to your running Chrome over CDP, so the agent sees your open tabs, cookies, extensions and logins. Chrome asks once for remote-debugging permission.

What about CI, or a server with no Chrome?

Run browser-use auth login and the same commands open a Browser Use cloud browser instead. Cloud browsers cost $0.02 per browser-hour and come with stealth, CAPTCHA solving and residential proxies.

Which agents?

Claude Code, Codex, Cursor, Gemini CLI, Pi, OpenClaw, and Hermes Agent, where it is the default browser driver. Anything that can run a shell command and read a skill file.

How is this different from Playwright MCP?

Playwright MCP gives the agent a fixed set of tools on a fresh browser. Here the agent writes the exact Python it needs against your real, logged-in browser, and the same commands work on cloud stealth browsers when you need scale.

Is there an MCP server too?

Two. browser-harness-mcp exposes the local browser helpers over stdio. The hosted server at api.browser-use.com/v3/mcp runs whole tasks in the cloud from Claude Code, Claude Desktop, Cursor or Windsurf.