Cloudflare Playwright

You run a Cloudflare Workers backed Model Control Protocol (MCP) server that lets an AI assistant automate a browser through Playwright. This server enables navigation, typing, clicking, and taking screenshots, all controlled by tools exposed through an MCP interface for compatible AI platforms.

How to use

You connect an MCP client to the Cloudflare Playwright MCP Server to perform browser automation tasks. Use the standard MCP URL endpoint to establish a connection, then invoke the available tools to navigate pages, interact with elements, type text, and capture screenshots. Keep your instructions concise and focused on a single action per step to get the most reliable results.

How to install

Prerequisites: you need Node.js installed on your machine and access to Cloudflare Wrangler for deployment.

# Install dependencies for the project
npm ci

Deploy the MCP server to Cloudflare Workers so it becomes accessible via an HTTP MCP URL.

# Deploy to Cloudflare Workers
npx wrangler deploy

Additional configuration and notes

You can connect using an MCP URL in the form https://<your-workers-subdomain>.workers.dev/sse. This URL is used by clients such as Cloudflare AI Playground and VS Code Copilot to request and execute browser actions.

For local or project-specific integration, you can also configure a local proxy or editor integration to reach the same MCP URL. One common approach is to route requests through a local proxy tool that forwards to the remote MCP endpoint.

Example tools exposed by the server include browser navigation, typing, taking screenshots, and interacting with page elements. Use simple, explicit prompts to trigger each tool.

Security and usage notes

Treat your MCP endpoint as you would any remote automation capability. Restrict access to trusted clients, rotate credentials if applicable, and monitor tool usage to prevent automated abuse.

Available tools

browser_navigate

Navigate the browser to a specified URL and update the page state.

browser_type

Type text into a focused element, with optional delays and keyboard actions.

browser_take_screenshot

Capture a screenshot of the current page and return the image in PNG format.

browser_clickt

Click a target element, optionally specifying coordinates or a selector.