Playwright MCP

Playwright MCP is Microsoft's MCP server for browser automation. It lets your agent drive a real browser through Playwright, reading pages as structured accessibility snapshots rather than screenshots, so it doesn't need a vision model and its actions are precise. What you can do with it - Navigate pages, click, type, hover, drag and select options - Fill out forms and upload files - Check that text, elements and values appear on the page - Take screenshots, and read console messages and network requests - Manage tabs, handle dialogs and run code in the page - Turn on more when you need it: cookies and storage, network routing, saving PDFs, tracing and video, and clicking by coordinates How it connects It runs on your machine with npx @playwright/mcp@latest and needs Node.js 18 or newer. There's also a Docker image, which runs headless Chromium only. Your browser profile By default it keeps a persistent browser profile for each project, so sign-ins carry over between sessions. Use --isolated for a fresh profile every session, or --extension to connect to your own running Chrome or Edge through the Playwright Extension. Good to know Playwright MCP isn't a security boundary: your agent can do anything in that browser. For coding agents, Microsoft suggests Playwright CLI with skills, which uses less of the agent's context.