This project provides an AI-driven browser automation server implementing the Model Context Protocol (MCP) for natural language control and web research. It targets developers and researchers building AI agents that require web interaction capabilities, offering seamless integration with various LLM providers and advanced browser automation tools.

How It Works

The server acts as an intermediary, translating natural language commands into browser actions via Playwright. It supports the MCP protocol for communication with AI agents, allowing them to execute tasks like page navigation, form filling, and element interaction. A key feature is its ability to connect to a user's existing Chrome/Chromium instance via the Chrome DevTools Protocol (CDP), enabling direct observation and use of personal browser profiles.

Quick Start & Requirements

• Install with uv sync.
• Requires Python 3.11+ and Playwright browsers (uv run playwright install).
• Configuration is primarily via environment variables (e.g., OPENROUTER_API_KEY).
• See Claude Desktop config example for setup.

Highlighted Details

• Full MCP protocol implementation for AI agent communication.
• Supports numerous LLM providers including OpenAI, Anthropic, Ollama, and OpenRouter.
• Offers run_browser_agent for natural language browser control and run_deep_search for multi-step web research.
• Can connect to a user's Chrome/Chromium instance via CDP (MCP_USE_OWN_BROWSER=true).

Maintenance & Community

No specific community links or notable contributors are mentioned in the README.

Licensing & Compatibility

MIT License. Permissive for commercial use and closed-source linking.

Limitations & Caveats

The project relies heavily on environment variables for configuration, which can be cumbersome for complex setups. While it supports connecting to an existing browser via CDP, managing browser state and potential conflicts requires careful user attention. Vision capabilities (MCP_USE_VISION) depend on the LLM model's support.