executeautomation/mcp-playwright

This project is a server that lets AI assistants like Claude control a real web browser. It connects to AI tools using a protocol called MCP (Model Context Protocol), which is a standard way for AI systems to call external tools. Once connected, the AI can open web pages, click buttons, fill in forms, take screenshots, scrape page content, and run JavaScript, all without a human touching the keyboard. The practical use case is automating repetitive browser tasks by typing instructions to an AI rather than writing automation scripts yourself. You might use it to test a website by telling Claude to visit a page and check that a button works, or to extract data from a page by asking the AI to read specific content. It works inside Claude Desktop, VS Code with GitHub Copilot, Cursor IDE, and similar AI-powered coding environments. Version 1.0.10 added device emulation across 143 preset device profiles. You can ask the AI to test a page as if it were an iPhone 13, an iPad, a Pixel phone, or a desktop browser, and the server adjusts the viewport size, user-agent string, touch support, and pixel ratio automatically to match that device. Installing the server requires running a single npm command. Browser binaries (Chromium, Firefox, or WebKit) download automatically the first time they are needed, so no separate browser installation step is required. Configuration involves adding a short JSON block to your AI tool's settings file that tells it where to find the server. The README includes separate config examples for Claude Desktop, VS Code, and HTTP mode for remote or headless deployments. The project is written in TypeScript and hosted on npm. Full documentation including a list of supported tools and a device quick-reference guide lives on the project's documentation site. The README is shorter than the full documentation, so some details are available only there.