stealth-browser

Stealth Browser MCP provides an undetectable browser automation layer you can control via any MCP client. The server exposes a rich set of tools organized into sections to bypass protections like Cloudflare, clone UI elements precisely, intercept and inspect network traffic, and drive real browser instances through the Chrome DevTools Protocol. With a modular architecture, you can run the full 90-tool arsenal or a lean 22-tool core, depending on your needs. To use it, start the Python server (for example by running python src/server.py) and connect your MCP client to the Stealth Browser MCP entry. You can then issue tasks such as navigating to pages, cloning page elements with pixel-accuracy, intercepting requests, executing CDP commands, and simulating human-like typing and input. The server supports running with flags like --minimal or --disable-* to tailor the toolset to your workflow. The toolbox is organized into sections such as browser-management, element-interaction, element-extraction, network-debugging, cdp-functions, dynamic-hooks, and more, making it straightforward to compose complex automation sequences for AI agents.

Prerequisites:\n- Python 3.8+ (recommended)\n- Git\n- (optional) Virtual environment utility, e.g., python -m venv\n\nStep-by-step installation:\n1) Clone the repository:\nbash\ngit clone https://github.com/vibheksoni/stealth-browser-mcp.git\ncd stealth-browser-mcp\n\n2) Create and activate a virtual environment:\nOn Windows:\nbash\npython -m venv venv\nvenv\Scripts\activate\n\nOn macOS/Linux:\nbash\npython3 -m venv venv\nsource venv/bin/activate\n\n3) Install Python dependencies:\nbash\npip install -r requirements.txt\n\n4) Run the server:\nbash\npython src/server.py\n\nNotes:\n- If you customize paths, ensure the working directory contains src/server.py.\n- For production or CI/CD, consider using virtual environments per project and pinning package versions.\n- The README also covers optional integration steps with Claude/fastmcp if you want to register the MCP server in clients.

Tips and common issues:\n- Ensure Python is on your PATH and you’re using a compatible version with requirements.txt.\n- When running behind strict proxies or VPNs, network interruptions can affect browser automation; verify network access and Cloudflare-related behavior separately during testing.\n- If you enable --minimal, verify that essential browser operations still meet your automation needs.\n- For MCP integration, you may provide an explicit env block in your MCP client config; currently the server presumes the default environment.\n- Debug by inspecting server logs for CDP command errors or element extraction issues; modern websites may require additional cloning strategies.\n- Docker or other containerized deployments are supported via standard MCP config, but you may need to expose required ports and mount the project directory correctly.