---
id: browser-harness
version: "0.1.8"
license: MIT
license_treatment: permissive
maintenance: active
---

# browser-harness — The simplest, thinnest, and most powerful harness to control your real browser with your agent.

License: permissive · Maintenance: active · Downloads: 7.1M/mo

## What it is and what it does

Browser Harness is a thin CDP harness that lets an LLM agent control your real browser directly, with one websocket connection to Chrome and no middleware. The agent can write and edit helper code and domain-specific skills during execution, improving the harness on each run. It's designed for tasks where you need complete freedom—the agent discovers what's missing and adds it.

The package ships with a command-line interface, stores agent-editable helpers and skills in your config directory, and supports both local Chrome instances (via --remote-debugging-port) and Browser Use Cloud browsers for stealth, sub-agents, or headless deployment. It depends on cdp-use, fetch-use, pillow, and websockets to handle Chrome protocol communication, image capture, and real-time bidirectional messaging.

Use it for:

- Let an LLM agent automate repetitive web tasks (file uploads, form filling, data entry) while it learns site-specific selectors and flows.
- Build domain skills for specific websites that the agent generates and improves iteratively.
- Deploy headless browser automation using Browser Use Cloud without managing local Chrome instances.
- Debug agent behavior by inspecting live browser state and letting the agent write custom helpers on the fly.
- Integrate browser control into multi-step AI workflows where the agent needs to interact with web applications dynamically.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Connects an LLM agent directly to a real Chrome/Chromium browser via the Chrome DevTools Protocol, allowing the agent to control browser tasks and self-improve by writing helper code during execution.

Yes, if you need an LLM to control a real browser and want the agent to improve itself during execution. The package is actively maintained, has low install friction, carries a permissive MIT license, and is in early alpha (0.1.8) with no known vulnerabilities. Start with the setup prompt in the description or Browser Use Cloud if you want to avoid local Chrome setup. Not suitable if you need a stable, production-hardened API—the alpha status and self-modifying helper code mean workflows may shift as the agent learns.

## Install

pip install browser-harness
uv add browser-harness
poetry add browser-harness

## Installing browser-harness

Before you install: Low friction: pure Python wheel with four runtime dependencies (cdp-use, fetch-use, pillow, websockets). Active maintenance—released 19 days ago with commits through 2026-08-14. Requires Python 3.11 or later.

License in practice: MIT license (permissive): you can use, modify, and distribute this package freely in commercial or private projects, with minimal restrictions.

Quickstart:

pip install browser-harness

from browser_harness import connect

# Connect to Chrome at default CDP endpoint
browser = connect()
# Agent can now control the browser and write helpers as needed

Requires Chrome or Chromium running with --remote-debugging-port enabled, or a Browser Use Cloud account; Python 3.11 or later.

Verify before relying:

- Repository star count and its relevance to package stability
- Whether the package works with headless Chrome or requires a display
- Performance characteristics when controlling complex web applications
- How agent-generated helper code persists and is versioned across runs

## Package facts

- License: MIT (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 7.1M/month (top 5,000 on PyPI)
- Known vulnerabilities: none known

## Tags

llm browser automation, agent browser control, chrome devtools protocol harness, ai web automation, browser agent framework, agent-automation, browser-control, llm-integration

[View on SkillFed](https://skillfed.io/packages/browser-harness) · [View on PyPI](https://pypi.org/project/browser-harness/)