Playwright MCP Server
The Playwright MCP Server lets the AI control a real browser: open pages, click buttons, fill forms, take screenshots, and extract content. You describe the task in natural language and it drives the browser to do it.
For example, "open this page, log in, and export the order list" — the AI works through it step by step via Playwright. This is Microsoft's official server.
What it does for you
Lets your AI actually drive a real browser — click, fill forms, screenshot, and scrape JS-rendered pages, including sites behind login.
Capabilities
- Open pages, click / type / submit
- Screenshot and read rendered page content
- Handle JS-rendered dynamic sites
- Multi-step browser automation
Once installed, try asking your AI
Say this directly in Claude / Cursor or any client with this server connected
- ›“Open this product page and tell me the price and stock status”
- ›“Fill out the signup form on this site and screenshot the confirmation”
Install / Connect
claude mcp add mcp -- npx -y @playwright/mcpFollow the official README; env and args vary per server.
Works with these clients
- Claude Desktopnpx
- Claude Codenpx
- Cursornpx
- VS Codenpx
- Windsurfnpx
Derived from the registry manifest (package type and transport) — we have not installed it ourselves. Check the official README before relying on it.
Why use it
Good for: developers and QA doing end-to-end testing, web automation, or data scraping.
Maintained by Microsoft with a very high TrustScore; full health data is shown below.
Similar / alternative servers
playwriter
Chrome extension & CLI to let agents control your browser. Runs Playwright snipp
@modelcontextprotocol/server-fetch
Model Context Protocol Servers
io.github.xberg-io/crawlberg
Crawlberg
firecrawl-mcp-server
🔥 Official Firecrawl MCP Server - Adds powerful web scraping and search to Curs
Frequently Asked Questions
Does it run a real browser or headless?
Both are supported and configurable. Headless is common for automation; run headed when you want to watch what's happening while debugging.
Can it get past login walls?
It operates a browser like a human, so it can go through normal login flows — but respect the target site's terms and don't use it for abusive scraping.