Playwright (Microsoft)

This is Microsoft's official Playwright MCP, and it's the most complete browser automation server you'll find for Claude. It gives you full programmatic control over Chromium, Firefox, and WebKit—launch browsers, navigate pages, click elements, fill forms, take screenshots, and extract content. The accessibility tree integration is the standout: Claude can query the DOM by role and label, which means you can say "click the Sign In button" instead of wrestling with CSS selectors. That's genuinely useful when you're prototyping scrapers or testing flows. I've used this for building quick QA scripts where Claude writes the test, runs it in a headless browser, and reports back with a screenshot if something fails. It's faster than writing Playwright code by hand for one-off tasks. The click-by-text feature works well for simple UIs but can get confused on dense pages with duplicate labels—you'll need to fall back to selectors occasionally. It handles modern SPAs without complaint, waits for network idle by default, and the screenshot output is clean enough to debug with. The setup is trivial if you're on Claude Desktop: one npx command, add three lines to your config, restart. Other hosts like Cursor or Cline need manual stdio wiring, which isn't hard but isn't documented in the repo. The server itself is stable—I haven't seen it crash or hang, even on heavy pages. Performance is what you'd expect from Playwright: launching a browser takes a second or two, navigation is near-instant after that. You shouldn't bother if you're just scraping static HTML—curl and a parser will be faster and lighter. This is for dynamic sites, SPAs, or anything that needs JavaScript execution. It's also overkill if you only need to fetch a page once; the value is in chaining actions (log in, navigate, click, extract). If you're already comfortable writing Playwright scripts, this won't replace your workflow—it's for when you want Claude to write and run the script for you.