browser-use

browser-use is an open-source Python library that makes websites accessible to AI agents. It lets an agent drive a real browser the way a human does, opening pages, clicking buttons, typing, and filling in forms, so a task described in natural language can be completed end to end. Rechecked 2026-08-21: the first-party repo is github.com/browser-use/browser-use, MIT licensed, with 109,924 stars and 3,454+ forks, created 2024-10-31. It is one of the most-starred AI agent infrastructure projects.

Built on Playwright, browser-use connects a large language model to a live Chromium browser. It is used both as a standalone automation layer and as a skill for agent CLIs such as Claude Code, Codex, Cursor, Hermes, and OpenClaw.

Key Features

• Human-like browsing: the agent opens pages, clicks, types, and fills forms just like a person.
• Natural-language tasks: describe what to do, such as "fill in this job application" or "extract my followers to a CSV," and it completes the task.
• Form filling and data extraction: structured data can be scraped and exported, for example to CSV.
• Skill integration: install the browser-use skill so agent CLIs (Claude Code, Codex, Cursor, Hermes, OpenClaw) can set it up automatically.
• Cloud option: Browser Use Cloud handles hosting and scaling when you do not want to run it yourself.

Use Cases

• AI agents that need to interact with real websites for automation and testing.
• Scraping and extraction: pull structured data from pages and export it.
• Workflow automation: agents that fill forms, apply to jobs, or manage accounts online.

Pricing

The core library is free and open source; the Cloud product is priced separately on browser-use.com.

Advantages

• Huge adoption: 109k+ stars make it a de facto standard for browser automation in agent stacks.
• Real browser: drives actual Chromium rather than a fake DOM, so results reflect what a user sees.
• Agent-ready: works as a skill inside popular coding-agent CLIs.

Getting Started

1. Install with uv on Python 3.12 (recommended), or pip install browser-use.
2. Connect the library to your browser.
3. Give an agent a task, or use it directly in Python to automate a page.
4. For agent CLIs, run browser-use skill install to register the skill.

See github.com/browser-use/browser-use and the docs.

Frequently Asked Questions

Is browser-use free?

The core library is open source under MIT. Browser Use Cloud is a separate paid product.

What browsers does it support?

It is built on Playwright, so it drives Chromium and other Playwright-supported browsers.

Can I use it inside Claude Code or Codex?

Yes. It ships as a skill that agent CLIs such as Claude Code, Codex, Cursor, Hermes, and OpenClaw can install.

Alternatives

• Playwright: the underlying multi-browser automation library.
• Chrome DevTools MCP: Google's MCP server that gives agents a live Chrome.
• Kitesurf: Cloudflare's hosted browser for agents.

Tips

1. Prefer uv with Python 3.12 for a clean install.
2. Use the skill installer inside your agent CLI to avoid manual setup.
3. For production workloads, evaluate Browser Use Cloud instead of self-hosting.

Conclusion

If you are building agents that need to actually use the web, browser-use is the de facto open-source layer to start with. Begin at github.com/browser-use/browser-use.