The Puppeteer MCP runs as a **local stdio server** with full browser control: - **Local Browser Instance**: Launches a real Chrome/Chromium browser on your machine - **Full Browser Access**: Can access any site, including those with stored cookies/sessions - **JavaScript Execution**: The evaluate tool can run arbitrary JavaScript in page contexts - **Screenshot Capture**: Can capture visual content of any page you visit **Security best practices**: Avoid automating sites where you're logged in with sensitive accounts, be cautious with the JavaScript evaluation feature, and consider that automated browsing may trigger security measures on some sites.

To start using the Puppeteer MCP: 1. **Add the server**: Configure your AI client with `npx -y @modelcontextprotocol/server-puppeteer` 2. **Chrome will launch automatically**: A browser instance starts when needed 3. **Start automating**: Ask your AI to navigate websites, fill forms, or take screenshots **Example prompts**: - "Navigate to example.com and take a screenshot" - "Fill out the contact form on this page" - "Click the submit button and wait for the confirmation" Puppeteer handles all the browser complexity—just describe what you want to accomplish.