Playwright

You can run a Playwright-based MCP server that exposes browser automation tools to AI assistants. This enables secure, scripted control of a browser for tasks like navigating sites, typing, clicking, capturing screenshots, and more, all through a Model Control Protocol interface.

How to use

Connect your MCP client to the server URL and start issuing actions through the available tools. You can deploy the server to Cloudflare Workers and then interact with it from compatible MCP clients such as Cloudflare AI Playground, Claude Desktop (via remote proxy), or VS Code integrations. Common workflows include navigating to a page, typing into inputs, clicking elements, and taking screenshots to verify results.

How to install

Prerequisites you need before installation:

• Node.js and npm installed on your machine
• Access to the internet to install dependencies
• A Cloudflare account for deploying to Workers (optional if you deploy to another environment)

Install dependencies for the project and prepare for deployment with these commands:

npm ci

Configure and deploy the MCP server

To deploy the MCP server to Cloudflare Workers, run the deployment command shown here.

npx wrangler deploy

Optional usage with Cloudflare AI Playground

Use a test URL to connect the MCP server with Cloudflare AI Playground. In the MCP Servers section, set the URL to the server’s SSE endpoint and connect. The playground will report available tools once connected.

VS Code and Claude Desktop setup (highlights)

If you want to use VS Code with the MCP server, you can add the server by URL to your environment and use it with Copilot-enabled tooling. For Claude Desktop, you can proxy a remote MCP server locally by configuring a local mcp-remote entry.

Available tools

browser_navigate

Navigate the browser to a specified URL and load the target page.

browser_type

Type text into an input field or contenteditable area within the page.

browser_clickt

Click a target element, such as a button or checkbox, identified by a selector or visible text.

browser_take_screenshot

Capture a screenshot of the current page state and return the image data.