Browser Use

What the browser-use MCP server does

The browser-use MCP server gives an AI agent access to browser automation for tasks that require visual or interactive work on the web. It can open pages, follow navigation flows, enter information into forms, select dates and times, retrieve page data, and complete multi-step tasks such as booking a driving test or finding information in a repository.

The project is aimed at agents that need to operate websites as a person would. That makes it a fit for sites without a suitable public API, workflows that depend on browser state, and tasks involving several pages or interaction steps.

How it works

The browser-use MCP server acts as a browser driver for an AI agent. Browser actions are provided to the agent, which determines the next interaction from the current page and the requested task. The underlying browser can run locally, use Browser Use Cloud, or connect to an existing remote browser through CDP in the documented integrations.

Browser Use also exposes an open-source agent for Python and supports TypeScript. Its documented Python path uses an LLM, an agent instance, and a local or cloud browser. The project also documents an Anthropic SDK integration that supplies browser actions and can add Bash for processing data or writing files on the SDK host.

Setup and configuration

The browser-use MCP server's broader project setup depends on the execution path. The Python library requires Python 3.11 or newer and can be installed with uv add browser-use. The example configuration uses OPENAI_API_KEY; BROWSER_USE_API_KEY is needed when using the BU2 model or Browser Use Cloud. The documented Anthropic integration requires an Anthropic SDK version that includes the browser tool, and Bash support requires Linux or macOS, or WSL on Windows.

The repository also documents a CLI workflow for installing or upgrading Browser Use with uv, registering its skill, and connecting it to a browser. That path names Claude Code, Codex, Hermes, and OpenClaw as possible agent environments.

Tools and capabilities

The browser-use MCP server supports the browser action set used by Browser Use. The repository states that its Anthropic integration implements all 31 browser actions, though the excerpt does not enumerate their individual names. Supported task categories include:

• Navigating websites and following multi-step flows
• Filling out web forms
• Selecting dates and times
• Extracting information from pages
• Handling CAPTCHA steps in demonstrated workflows
• Using local, cloud, or remote browser targets, depending on the integration

The project can also pair browser control with Bash in its Anthropic SDK integration for local data processing and file output.

Limitations and notes

Browser automation depends on the target site's layout, access controls, login state, and other runtime conditions. The provided material demonstrates CAPTCHA handling but does not define universal site coverage or guaranteed task completion. Cloud browsers and the BU2 model require a Browser Use API key, while the local Python example uses an OpenAI API key. Bash integration is limited to hosts with /bin/bash, with WSL recommended for Windows users.

The excerpt does not list individual MCP tool names, transport settings, or a dedicated MCP installation command. Consult the repository and linked documentation for those details before deploying the browser-use MCP server.