Puppeteer MCP

What it does

Puppeteer MCP wraps the official Puppeteer library — Google's Node.js API for controlling headless Chrome. Claude gets a real Chromium browser instance, enabling it to interact with any web page exactly as a user would, including full JavaScript execution, cookie handling, and dynamic content rendering.

It is part of the official Anthropic MCP reference server collection, making it a reliable choice for tasks like taking visual screenshots, testing web applications, automating repetitive browser tasks, or scraping content from JavaScript-dependent pages.

• Navigate to any URL with full JS execution
• Take full-page or viewport screenshots
• Click elements, type text, and submit forms
• Evaluate arbitrary JavaScript in the page context
• Extract HTML content from rendered pages
• Wait for specific selectors before proceeding

How to connect

claude mcp add puppeteer -- npx -y @modelcontextprotocol/server-puppeteer

Available tools

• navigate — go to a URL and wait for page load
• screenshot — capture a PNG screenshot of the current page
• click — click an element by CSS selector
• type — type text into a focused element
• evaluate — execute JavaScript in the page context
• get_content — return the page's HTML content
• wait_for_selector — wait until a selector is visible

Example usage

Ask Claude to take a screenshot of a competitor's pricing page or extract rendered text from a SPA:

# User prompt
Take a full-page screenshot of github.com and
describe the layout and key UI elements you see.

navigate("https://github.com")
wait_for_selector(".application-main")
screenshot({ fullPage: true })
→ returns base64 PNG, ~1200x3400px
→ Claude describes layout elements