browser-use/browser-harness
Browser Harness | Self-healing harness that enables LLMs to complete any task.
Browser Harness by browser-use is a minimal Chrome DevTools Protocol framework that gives LLM agents a direct websocket to your real Chrome browser. Self-healing helpers mean the agent writes missing utilities during execution. Built for tasks where you need complete browser freedom.
Our Review
Browser Harness is the browser-use team's minimal CDP-based framework for letting an LLM agent drive a real Chrome browser with full freedom -- 13,100 stars and 1,200 forks in under a month as of May 2026. Released April 2026 by browser-use under MIT and distinct from the larger browser-use library, it ships a thin one-websocket-to-Chrome design where the agent writes the missing helpers itself during execution.
What Browser Harness does:
- •One websocket to Chrome a minimal Chrome DevTools Protocol layer with nothing between the agent and the real browser; no Playwright abstraction overhead.
- •Self-healing helpers when the agent needs a helper that does not exist (file upload, custom form interaction), it writes one into agent_helpers.py and uses it.
- •Real browser, not Chromium-only attaches to your own Chrome via chrome://inspect/#remote-debugging, so cookies, extensions, and your login state are all available.
- •Domain-skill examples agent-workspace/domain-skills/ ships example tasks the agent can adapt for new sites.
- •Agent-first install the setup prompt is designed to be pasted into Claude Code or Codex, which then runs the install itself by reading install.md.
- •Telegram + VPS deployment topics include telegram-agent and vps-agent, so the same harness can run unattended on a remote machine.
- •Compatible with Browser Use Cloud works alongside browser-use's broader cloud stack for stealth, proxies, and managed scaling.
Browser Harness ecosystem:
- •browser-use/browser-use the larger library covering more abstracted agent patterns with Playwright underneath.
- •Browser Use Cloud managed hosting for harnessed agents with proxy rotation and captcha solving.
- •Discord community shared with the main browser-use community.
- •CDP / Chrome DevTools Protocol the underlying protocol the harness wraps directly.
Getting started:
Paste this into Claude Code or Codex: "Set up https://github.com/browser-use/browser-harness for me. Read install.md and follow the steps to install browser-harness and connect it to my browser." The agent will open chrome://inspect/#remote-debugging, prompt you to tick the checkbox, and finish the install. Chrome 144+ adds a per-attach popup you must click Allow on.
Limitations:
Attaches to your real Chrome via remote debugging -- this is a security boundary you should understand before granting access. Cookies, sessions, and tab state are visible to the agent. Requires Python and an LLM API key with budget; harness self-healing writes new code per task, costing tokens. Project is brand new (April 2026), so expect rapid changes. Designed for tasks where you want maximum freedom, not for production workflows that need strict guardrails.
Our Verdict
Browser Harness in 2026 is the most minimal LLM-to-real-browser framework on GitHub. 13,100 stars in under a month, 1,200 forks, MIT licensed, and built by the same browser-use team behind the larger browser-use library. The pitch -- "one websocket to Chrome, nothing between" -- captures the design philosophy.
For Vibe Builders, Browser Harness is the right tool when you need an agent to do something on a website that no off-the-shelf scraper handles. The self-healing helper pattern means the agent writes whatever code is missing, so you do not pre-script every interaction. The trade-off is the agent runs against your real Chrome with your real cookies.
For Developers building agent-driven RPA workflows, the CDP-direct design is faster than Playwright-wrapped alternatives and easier to extend. The agent_helpers.py self-modification pattern is interesting research material even if you do not adopt the full framework.
Skip Browser Harness if you need strict sandboxing -- this attaches to your real Chrome by design. Skip if you want a high-level scraper API; the larger browser-use library is more polished for that. For locked-down enterprise RPA, UiPath or Microsoft Power Automate cover that ground with proper audit trails and role-based access.
Frequently Asked Questions
What is Browser Harness?
Browser Harness is browser-use's minimal Chrome DevTools Protocol framework for letting an LLM agent drive a real Chrome browser. Released April 2026 by browser-use under MIT, it ships a one-websocket-to-Chrome design where the agent writes missing helper functions itself during execution. As of May 2026 it has 13,100 GitHub stars and 1,200 forks in under a month.
How is it different from browser-use/browser-use?
browser-use/browser-use is the larger Playwright-based library covering more abstracted agent patterns with built-in domain logic. Browser Harness is intentionally minimal: one CDP websocket, no Playwright wrapper, and the agent writes its own helpers. Use Browser Harness when you want maximum freedom; use browser-use when you want pre-built abstractions and faster setup.
How does the self-healing harness work?
When the agent encounters a task it does not have a helper for (uploading a file, interacting with a custom dropdown, etc.), it writes the missing helper into agent_helpers.py during the run and then uses it. The harness improves with each run as helpers accumulate. The trade-off is each new task type costs tokens to generate the helper, but reuse is free.
How do I install Browser Harness?
The recommended path is paste-driven. Open Claude Code or Codex in the repo and paste 'Set up https://github.com/browser-use/browser-harness for me. Read install.md and follow the steps.' The agent will open chrome://inspect/#remote-debugging, prompt you to tick the remote-debugging checkbox, and finish the install. On Chrome 144+ a per-attach popup will appear; click Allow.
Is Browser Harness safe to run?
It attaches to your real Chrome via remote debugging, which exposes cookies, sessions, and tab state to the agent. This is by design for maximum freedom but means you should not run it on a machine with sensitive logged-in accounts unless you trust the agent and the LLM provider. For production isolated workflows, use Browser Use Cloud or a sandboxed alternative instead.
What is browser-harness?
Browser Harness by browser-use is a minimal Chrome DevTools Protocol framework that gives LLM agents a direct websocket to your real Chrome browser. Self-healing helpers mean the agent writes missing utilities during execution. Built for tasks where you need complete browser freedom.
What license does browser-harness use?
browser-harness uses the MIT license.
What are alternatives to browser-harness?
Explore related tools and alternatives on My AI Guide.
Open source & community-verified
MIT licensed: free to use in any project, no strings attached. 13,344 developers have starred this, meaning the community has reviewed and trusted it.
Reviewed by My AI Guide for relevance, quality, and active maintenance before listing.
Topics