MCP Server Playwright

What automatalabs/mcp-server-playwright MCP server does

The automatalabs/mcp-server-playwright MCP server gives an MCP client access to browser automation through Playwright. It supports opening URLs, taking screenshots, clicking controls, hovering over page elements, entering form values, choosing options in HTML SELECT elements, and evaluating JavaScript in the browser.

Interaction tools accept either CSS selectors or visible text for several operations. This lets an agent target a known element such as #submit-button, or use text such as a button label when a selector is not available. Screenshots can cover the entire page or a selected element and are exposed through a screenshot resource after capture.

The server also provides browser console output through the console://logs resource. That can help an agent inspect messages produced while a page is running.

How it works

The package runs as an MCP server over a local stdio configuration. An MCP client starts the package with npx, and the server exposes Playwright actions as named tools. Browser operations are performed in a real browser environment, while results such as captured images and console output are made available through MCP resources.

Available tools include browser_navigate, browser_screenshot, browser_click, browser_click_text, browser_hover, browser_hover_text, browser_fill, browser_select, browser_select_text, and browser_evaluate. Screenshot resources use the name supplied when the screenshot is created.

Setup and configuration

The automatalabs/mcp-server-playwright MCP server requires Node.js with npx available on the system PATH. The documented client configuration starts the package with npx -y @automatalabs/mcp-server-playwright.

For Claude Desktop, the package installation flow can create or update the client configuration on Windows and macOS. The resulting MCP entry uses npx as the command and passes -y plus the npm package name as arguments. Cursor is also documented as a supported client. Its setup includes installing Playwright browsers with npx playwright install before adding the MCP server configuration.

The repository also documents installation through Smithery and mcp-get, but those are setup methods rather than the server process itself.

Tools and capabilities

• Navigate to any URL with browser_navigate.
• Capture a full-page or element screenshot with browser_screenshot.
• Click or hover using CSS selectors or text content.
• Fill input fields and select HTML SELECT options.
• Run JavaScript through browser_evaluate.
• Read browser console messages from console://logs.
• Retrieve PNG screenshots through screenshot://<n> using the capture name.

Limitations and notes

The provided material describes page interaction and browser-context JavaScript execution, but does not specify authentication helpers, persistent browser profiles, file upload handling, network interception, or a testing-specific assertion API. Those capabilities should not be assumed from this listing.

Selectors and text targets must identify the intended page elements. JavaScript passed to browser_evaluate runs in the browser context, so its behavior depends on the current page and its loaded state. The project is licensed under MIT.