Playwright MCP
by Microsoft
Microsoft's official MCP server that gives coding agents a real, controllable browser — driven through the accessibility tree instead of screenshots, for deterministic clicking, typing, and testing.
Add to Claude Code
claude mcp add playwright npx @playwright/mcp@latest
About
Playwright MCP operates on the accessibility tree — the structured representation of what's on a page — rather than screenshots and pixel coordinates, so an agent can click, type, fill forms, and navigate deterministically without a vision model in the loop.
Beyond testing, the same tool surface covers general browser automation: scraping structured data, generating PDFs, monitoring network requests, and evaluating JavaScript on a live page.
Key features
- Accessibility-tree-based interaction — no screenshots or pixel coordinates needed
- Click, type, fill forms, and navigate deterministically
- Take screenshots, generate PDFs, and capture network traffic
- Isolated browser profile by default, or connect to an existing Chrome instance
Use cases
- Verify a UI change actually works by driving the real browser end to end
- Scrape structured data from pages that require JavaScript rendering
- Automate repetitive browser workflows like form submission or data entry
Available tools
browser_navigate
Navigates the browser to a given URL.
browser_snapshot
Captures the current accessibility tree of the page.
browser_click
Clicks an element identified from the accessibility snapshot.
browser_type
Types text into a focused input field.
browser_screenshot
Takes a screenshot of the current page or a specific element.
Frequently asked questions
Does my agent see my logged-in browser?
Not by default — it launches an isolated browser profile. You can point it at a persistent profile or an existing Chrome instance via flags if you need an authenticated session.
What do I need installed?
Node.js 18 or later — Playwright installs its own browser binary automatically on first run.