Playwright MCP

What it does

Playwright MCP gives Claude a full, real browser — not a simplified HTTP fetcher. Claude launches a Chromium, Firefox, or WebKit instance, navigates to URLs, and interacts with pages exactly as a human user would. JavaScript executes, SPAs render, and dynamic content loads normally.

This makes it the right tool for any task requiring actual page interaction: testing login flows, automating form submissions, scraping content that requires JavaScript rendering, or taking visual screenshots for design review.

• Navigate to any URL and wait for full page load
• Click any element by CSS selector, text, or role
• Type text into inputs, textareas, and rich editors
• Take full-page or element-level screenshots
• Extract text content from any page or element
• Scroll, hover, and trigger mouse events
• Select dropdown options and interact with checkboxes
• Wait for elements to appear, disappear, or change
• Execute arbitrary JavaScript in the page context

How to connect

claude mcp add playwright -- npx -y @playwright/mcp

To use a headed (visible) browser instead of headless mode, pass the --headed flag:

claude mcp add playwright -- npx -y @playwright/mcp --headed

Available tools

• navigate — go to a URL and wait for load
• click — click an element by selector or text
• type — type text into a focused input
• screenshot — capture a screenshot of the page or element
• get_text — extract visible text from the page or a selector
• scroll — scroll the page or an element
• hover — hover over an element
• select_option — select a value from a dropdown
• wait_for_element — wait until a selector is visible
• evaluate_js — run JavaScript in the page context

Example usage

Ask Claude to test a login flow or audit a landing page visually:

# User prompt
Go to app.example.com/login, fill in the email and
password fields, click the submit button, and take
a screenshot of whatever page loads next.

navigate("https://app.example.com/login")
type("input[name=email]", "user@example.com")
type("input[name=password]", "••••••••")
click("button[type=submit]")
wait_for_element(".dashboard")
screenshot() → returns base64 image