Launch AI Works
InsightsNov 9, 20264 min read

Browser Harness: Let AI Use Your Chrome Browser

Browser Harness connects Claude Code, Codex, Cursor and other agents directly to Chrome via the DevTools Protocol, offering a low‑cost way for small businesses to automate browser‑based work.

LBy Launch AI Works
Browser Harness: Let AI Use Your Chrome Browser
The short version

Browser Harness is a MIT‑licensed bridge that lets AI agents control your real Chrome session with a single prompt. It costs less per task than comparable tools while delivering similar accuracy.

What Browser Harness Is

Browser Harness is an open‑source library (MIT license) that enables AI coding agents such as Claude Code, Codex or Cursor to operate your actual Chrome browser. The agent can open pages, click buttons, type into fields, read content and download files, all while you remain logged in with your own credentials. The project lives on GitHub and has attracted 19 000 stars and 1.9 k forks, indicating strong community interest.

How It Connects Your Browser to an AI Agent

The tool talks to Chrome over the Chrome DevTools Protocol (CDP). After a single setup prompt is pasted into the coding agent, the agent launches Chrome 144+ with remote‑debugging enabled (chrome://inspect). The user grants permission once, and the harness can then issue CDP commands to control any open tab or the whole browser. Because the connection uses the real browser profile, the AI works with the same cookies, extensions and logged‑in sessions you already have.

Browser Harness can be used with any MCP‑compatible client – Claude Code, Codex, Cursor, Copilot, Gemini and others. It also supports a cloud version (Browser Use Cloud) for teams that prefer not to run a local Chrome instance. The harness automatically writes helper scripts in the agent’s workspace when a missing function is needed, improving its capabilities over time.

Performance and Cost Compared to Alternatives

In a benchmark of 100 public Browser Use tasks, Browser Harness achieved a mean rubric score of 60.4, essentially on par with Playwright MCP (60.0) and Chrome DevTools MCP (61.0). The confidence intervals overlapped, meaning no tool showed a statistically significant advantage in accuracy.

Where Browser Harness stands out is efficiency. Its median input token count was 2.1 million, compared with 3.3 million for the agent‑browser baseline, 4.4 million for Playwright MCP and 6.7 million for Chrome DevTools MCP. Lower token usage translates to lower model cost: the median cost per task was $0.042, while the other tools ranged from $0.056 to $0.097. The time to complete a task was 9.0 minutes, slightly slower than Playwright MCP (8.3 min) and Chrome DevTools MCP (8.7 min) but faster than the agent‑browser baseline (11.2 min).

These numbers show that Browser Harness delivers comparable accuracy at the lowest token and monetary cost, making it a practical choice for budget‑conscious small businesses.

What This Means for Small Businesses

For a small business, the biggest pain point in web automation is handling authentication and session management. Browser Harness eliminates the need to recreate logins or export cookies because the AI works inside the Chrome profile you already use. This opens up several practical use cases:

• **Checkout testing** – An AI agent can run through your e‑commerce checkout flow using real payment credentials, catching broken steps before customers do.

• **Report retrieval** – Vendor portals often require manual navigation to download sales or inventory reports. The agent can log in, navigate, and save the files automatically.

• **Form filling** – Repetitive data entry tasks, such as updating product listings on multiple marketplaces, can be delegated to the AI without exposing passwords to a separate script.

Because the model cost per task is under five cents, running dozens of such automations each week remains affordable. The open‑source nature also means you can inspect the code, host it yourself, or contribute improvements without licensing fees.

Getting Started Quickly

To try Browser Harness, install the Python package (requires Python 3.12) and register the skill with the command `browser-harness skill`. Then paste the single setup prompt into your chosen AI agent. The agent will open chrome://inspect, ask you to enable remote debugging, and once you confirm, it can start controlling the browser. If the connection fails, the repository’s install guide provides step‑by‑step troubleshooting.

Because the tool works with both local Chrome and the hosted Browser Use Cloud, you can begin with a local test and later move to a cloud instance if you need a dedicated, isolated environment for higher‑volume automation.

Takeaways

  • Browser Harness lets AI agents act inside your real, logged‑in Chrome, removing the need for separate authentication scripts.
  • It achieves comparable task accuracy to Playwright and Chrome DevTools MCP while using the fewest tokens and costing as little as $0.042 per task.
  • The single‑prompt setup and open‑source MIT license make it a low‑overhead option for small businesses looking to automate web‑based workflows.
L

Launch AI Works builds practical AI and automation inside small businesses. Notes in Insights are written from sources we cite and checked before publishing.

Your turn

Where would this work in your business?

Take the two-minute AI Opportunity Snapshot and we will point to where automation is most likely to pay off for you.

2mFree AI Opportunity Snapshot →