Browser Harness is a browser automation framework that enables large language models to complete web tasks by connecting directly to a real browser through a Chrome DevTools Protocol websocket.
The tool solves the problem of LLMs needing reliable, adaptable ways to interact with web browsers. It works by establishing a direct connection to Chrome's debugging interface, allowing an agent to control the browser while simultaneously writing and improving helper functions as it encounters new tasks. This self-healing approach means the harness becomes more capable over time, with the agent creating reusable skills in its local workspace rather than modifying the core framework itself.
Developers should choose this tool if they want to automate personal or logged-in web workflows without building custom browser control logic. It suits individual use cases where an LLM agent needs to interact with your actual browser session, as well as scaled deployments through the associated Browser Use Cloud service, which adds features like parallel browser instances, live previews, proxies, and CAPTCHA solving. The tool integrates with MCP servers, making it compatible with Claude Code, Devin, Cursor, and other MCP clients that want to drive browser automation without implementing their own CDP layer.
The project maintains an active codebase with ongoing refinement of its core harness logic and documentation. The maintainers actively solicit contributions including bug fixes, documentation improvements, and agent-generated domain skills, indicating a collaborative development model. The project includes supplementary resources explaining the design philosophy behind agent harnesses and how web agents can learn from repeated interactions.
