explaingit

browser-use/browser-harness

12,497PythonAudience · vibe coderComplexity · 3/5Setup · moderate

TLDR

A thin bridge that connects an AI language model to your real Chrome browser via Chrome DevTools Protocol, letting the AI browse, click, and fill forms, and write missing automation helpers itself when it gets stuck, growing smarter with each use.

Mindmap

mindmap
  root((browser-harness))
    What it does
      AI controls Chrome
      Reads and clicks pages
      Writes its own helpers
    How it works
      Chrome DevTools Protocol
      Self-healing code writing
      Domain skill files
    Setup
      Paste prompt into Claude Code
      Enable Chrome debugging
      Click Allow popup
    Cloud tier
      Multiple browsers
      Proxy support
      CAPTCHA solving
Click or tap to explore — scroll the page freely

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

Things people build with this

USE CASE 1

Automate repetitive browser tasks like form filling or data extraction by describing the task to an AI instead of writing automation code.

USE CASE 2

Build reusable site-specific skill files for GitHub or LinkedIn that the AI reuses across future automation runs.

USE CASE 3

Use the cloud tier to run AI browser automation tasks without keeping a local Chrome instance open.

Tech stack

PythonChrome DevTools Protocol

Getting it running

Difficulty · moderate Time to first run · 30min

Requires Chrome with remote debugging enabled on port 9222, initial setup is guided by pasting a prompt into an AI coding tool rather than manual configuration.

No license information is mentioned in the explanation.

In plain English

Browser Harness connects an AI language model directly to your Chrome browser through a single protocol called Chrome DevTools Protocol (CDP). Instead of sitting between the AI and the browser with a rigid automation layer, this project acts as a thin bridge: the AI reads the page, decides what to do, and executes actions in your real browser rather than a sandboxed copy. The key idea is that when the AI encounters a task it does not know how to complete (like uploading a file or navigating a site-specific flow), it writes the missing helper code itself and saves it. The next time a similar task comes up, the saved helper is reused. The README describes this as a self-healing system: the AI fills in the gaps during actual use, making the setup more capable over time without you writing code manually. Setup is intentionally hands-off. You copy a short prompt into Claude Code or a similar AI coding tool, and the AI reads the project's install instructions and configures everything itself, including connecting to Chrome's remote debugging interface. You enable one checkbox in Chrome settings and click Allow when a popup appears. The project is roughly 1,000 lines spread across four main pieces: an install guide, a usage guide, a protected core package, and a workspace folder where the AI writes and updates its own helper code. A folder of domain skills holds site-specific playbooks the AI builds for places like GitHub, LinkedIn, and Amazon. Community contributors can submit their AI-generated skill files via pull request. Browser Use also offers a hosted cloud tier with free access to multiple concurrent browsers, proxy support, and CAPTCHA solving, for users who want to run tasks without keeping a local browser open.

Copy-paste prompts

Prompt 1
Using browser-harness, write a browser automation task that logs into GitHub, navigates to a specific repo, and opens a new issue with a given title and body.
Prompt 2
Walk me through setting up browser-harness with my local Chrome browser, including enabling remote debugging and connecting the AI agent.
Prompt 3
Using browser-harness's skill file system, create a reusable skill that navigates LinkedIn job search, filters by location, and extracts the top 10 job titles and company names.
Open on GitHub → Explain another repo

← browser-use on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.