Playwright MCP Server: Browser Automation with Claude and Cursor

The Playwright MCP server gives your AI assistant a real web browser it can control programmatically. That means filling forms, clicking buttons, taking screenshots, scraping dynamic pages, and running end-to-end tests — all from a single conversation with Claude or Cursor.

This guide covers what the Playwright MCP server can do, how to install it, and real-world workflows you'll use every day.

What is the Playwright MCP Server?

Playwright is Microsoft's battle-tested browser automation library, trusted by engineering teams worldwide for end-to-end testing. The Playwright MCP server wraps Playwright in a Model Context Protocol interface, giving AI assistants direct browser control.

Unlike simpler scraping tools, Playwright handles:

• JavaScript-heavy pages — SPAs, React/Vue apps, dynamic content
• Authentication flows — login forms, OAuth, cookie management
• Screenshots and PDFs — capture pages as images or documents
• Network interception — inspect and mock API responses
• Multi-browser support — Chromium, Firefox, and WebKit

Installation

Install the Playwright MCP server package:

npm install -g @playwright/mcp

Or use it via npx (no global install needed):

npx @playwright/mcp@latest

If you're using npx, Playwright will auto-download browser binaries on first run.

Configuration: Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%Claudeclaude_desktop_config.json (Windows):

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest"]
    }
  }
}

For headless mode (no visible browser window):

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest", "--headless"]
    }
  }
}

Configuration: Cursor

In ~/.cursor/mcp.json:

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest", "--headless"]
    }
  }
}

Available Tools

The Playwright MCP server exposes a rich set of browser interaction tools:

• browser_navigate — Navigate to any URL
• browser_click — Click elements by selector or description
• browser_type — Type text into inputs
• browser_screenshot — Capture page screenshots
• browser_get_text — Extract visible text from a page
• browser_get_html — Get full page HTML
• browser_scroll — Scroll the page
• browser_wait_for — Wait for selectors or text to appear
• browser_evaluate — Execute JavaScript in the page context
• browser_pdf — Generate a PDF from the current page

Real-World Use Cases

Web Scraping Dynamic Pages

Many sites require JavaScript execution before content appears. Ask Claude: "Go to hacker news, scrape the top 10 post titles and vote counts, and give me a summary."

End-to-End Testing

Ask: "Navigate to localhost:3000, log in with test@example.com / password123, verify the dashboard loads with 5 dashboard cards, and take a screenshot."

Competitor Research

Ask: "Visit our three main competitor pricing pages, take screenshots of their pricing tables, and compare their plans."

Form Automation

Ask: "Go to the contact form at example.com/contact, fill in the details for our partnership inquiry, review what you've filled in, and take a screenshot before submitting."

Visual Regression Detection

Ask: "Screenshot the homepage at 1280px width and 375px width (mobile). Are there any layout issues visible?"

Playwright vs Puppeteer MCP: Which to Use?

Both offer browser automation via MCP. The key differences:

• Playwright MCP supports Chromium, Firefox, and WebKit. Better cross-browser coverage. More reliable element selection.
• Puppeteer MCP is Chromium-only. Lighter weight. Good if you only need Chrome and want a smaller dependency footprint.

For most production use cases, Playwright is the better choice. For quick local automation, either works fine.

Headless vs Headed Mode

By default, Playwright opens a visible browser window (headed mode). This is useful for debugging — you can see exactly what the AI is doing. For automated tasks, use --headless to run without a visible window.

Related guides:

• Best MCP Servers for Automation
• Best MCP Servers for Testing
• Best MCP Servers for Developers