Skip to content
Get packs

Guide · Agent skill

How to use the Browser Use agent skill (live Chrome control)

Updated · written with AI from the project's own documentation and checked against our catalogue data

The browser-use skill tells an AI agent to work in your real, already-signed-in Chrome profile instead of a fresh test browser. It is useful when the page depends on your cookies, single sign-on, device trust or extensions. This guide explains what the skill does, how to install it, how to ask for it and what it needs before you rely on it.

Browser Use (Live Chrome Control)steipete/agent-scripts · MIT

Quick start

  1. 1Download and unzip. Download the browser-use ZIP from this page and unzip it into ~/.claude/skills/ or your project's .claude/skills/ folder.
  2. 2Set up OpenClaw. Run openclaw browser extension install, load the printed path with Load unpacked in Chrome, then check it with openclaw browser extension status --json.
  3. 3Check mcporter. Confirm the chrome-devtools server with mcporter config get chrome-devtools --json, then make sure your canonical ~/.mcporter/mcporter.json holds the Chrome definition.
  4. 4Verify the connection. Run mcporter call chrome-devtools.list_pages --args '{}' --output text. Success means the output lists your visible tabs.
  5. 5Ask for a live check. Make a request that needs your signed-in browser, such as checking a logged-in page, and watch that the relay is used rather than a fallback.

What does the browser-use skill make the agent do?

Agent skills are folders of instructions that an AI agent loads when a task matches them. This skill covers browser control: the agent is told to operate your existing Chrome profile, especially for login-dependent work and checking a rendered web UI.

The skill sets a routing order. It first uses a Codex Chrome plugin if that plugin is callable in the active session. Otherwise it prefers the OpenClaw extension-backed mcporter route. Direct DevTools attachment is only a labelled last fallback. The skill also includes a helper script, auto-allow.py, that watches for Chrome's Allow remote debugging prompt and presses Allow automatically.

  • Controls the Chrome profile you are signed into, not an isolated browser
  • Prefers a callable Chrome plugin, then the OpenClaw relay, then direct DevTools as a last resort
  • Includes auto-allow.py for the remote debugging prompt

How do you install the browser-use skill?

Download the browser-use ZIP from this page and unzip it. For a personal setup, place the browser-use folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

You can also upload the skill in the Claude apps. Open Settings, go to Capabilities, then Skills, and upload the ZIP there. Either way, the agent should pick up the skill from its SKILL.md file.

  • Personal: ~/.claude/skills/browser-use
  • Project: .claude/skills/browser-use inside your project
  • Claude apps: Settings → Capabilities → Skills

How do you set up OpenClaw and mcporter for the skill?

The preferred route depends on OpenClaw and its browser extension. Install the extension with the command below, then load it in Chrome using Load unpacked. Check the installation with the status command. Its output should report no issues and manualSetupRequired set to false.

mcporter reads its canonical config at ~/.mcporter/mcporter.json. The skill's mcporter-config.md file explains how to confirm the chrome-devtools server points at the right config. Verify the setup by listing your open tabs.

  • Restart Chrome once if the extension tried native messaging before it was installed
  • Confirm the popup shows Connected
  • Stop the mcporter daemon after changing the relay route, then repeat the test
openclaw browser extension install openclaw browser extension status --json mcporter config get chrome-devtools --json mcporter call chrome-devtools.list_pages --args '{}' --output text

How do you trigger the skill with natural requests?

You do not need special syntax. Ask for work that needs your logged-in browser, and name the page or the behaviour you want checked. The agent should match the request to the skill from its description.

Requests that fit the skill are ones involving login state or a live UI check. Requests that do not need your real profile, such as checking a public API with curl, are better handled without it.

  • Check the billing page in my Chrome and tell me whether the invoice total renders correctly.
  • Use my signed-in browser to open the admin dashboard and confirm the filter bar works.
  • Verify the layout bug on the staging page in my own Chrome, not a test browser.
  • List the tabs I have open right now in Chrome.

What should you check for safety?

The skill can act with your accounts and sessions, so treat it as access to your real browser. The skill says that Playwright, Puppeteer, AppleScript, osascript, generic GUI scripting, the macOS open command and isolated browsers should not be used as substitutes unless you explicitly ask for them.

The auto-allow.py helper presses Allow on Chrome's remote debugging prompt. That removes a security confirmation step, so avoid it when sensitive sessions such as banking, email or admin consoles are open. The skill also says credentials should stay out of configuration, command output, chat, logs and screenshots. The skill declares that it runs scripts and does not use the network.

  • Do not use it while banking, email or admin consoles are open
  • Expect an Allow remote debugging prompt or relay-policy error if the extension transport was not used
  • Restricted pages, incognito windows and other profiles are excluded

What are the requirements and tips?

You need Google Chrome, OpenClaw with its browser extension installed, and mcporter with a canonical ~/.mcporter/mcporter.json. auto-allow.py needs Python 3 and, on macOS, the accessibility tooling axorc. The skill mainly depends on the OpenClaw setup, so it is a poor fit without it.

Tips: test the local relay first, use Selected tabs mode with the Chrome tab group named OpenClaw if you want to limit exposure, and stop the daemon with mcporter daemon stop after relay changes. New pairings default to All tabs, so check the popup mode before a sensitive session.

  • Use mcporter daemon status --json to confirm the relay decision
  • Set chromeDevtoolsRelay to require in the canonical Chrome definition to enforce relay-only routing
  • Use a sufficient discovery timeout rather than a guessed port
MCPORTER_CHROME_DEVTOOLS_RELAY_POLICY=require mcporter call chrome-devtools.list_pages --timeout 300000 --output text

What people use it for

  • Verifying a rendered web UI bug in your own logged-in Chrome
  • Automating work on sites that need your existing login, SSO or device trust
  • Checking a staging page with your own cookies and extensions
  • Listing which tabs are open before a browser task

Frequently asked questions

How do I install the browser-use agent skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings, Capabilities, Skills.

Is the browser-use skill free?

The skill is published under the MIT licence. This page does not list any price, so check the source repository for licence terms before reuse.

What do I need to run the browser-use skill?

You need Google Chrome, OpenClaw with its browser extension installed, and mcporter with a canonical ~/.mcporter/mcporter.json. auto-allow.py also needs Python 3 and, on macOS, the axorc accessibility tooling.

Why does Chrome show an Allow remote debugging prompt?

That prompt means the extension transport was not used and the agent fell back to direct DevTools attachment. Stop the mcporter daemon, confirm the relay with the status commands, and repeat the test.

Does the skill use the network?

The skill's metadata says it does not use the network, but it does run scripts. Review auto-allow.py before using it, since it presses Allow on a security prompt.

How is this different from Playwright or Puppeteer?

Playwright and Puppeteer usually start a separate browser. This skill is built to control your existing signed-in Chrome, and it says those tools should not be used as substitutes unless you explicitly ask.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you need an agent to work in your own signed-in Chrome, especially for live UI checks, and you already run OpenClaw and mcporter. Pick a different approach if you do not want an agent acting with your personal accounts, or if you work in sensitive sessions where an automatic Allow press is unacceptable.

Download the skill