Cloudflare Playwright

This MCP server lets you control a browser through a defined set of tools, enabling web automation tasks such as navigation, typing, clicking, and taking screenshots. It is designed to work with AI assistants through standard MCP clients, letting you perform complex browser actions in an automated, chat-like workflow.

How to use

Connect your MCP client to the Cloudflare Playwright MCP server to start browser automation. You will typically use an HTTP MCP URL for remote access or a local Proxy/CLI setup that forwards requests to the cloud service.

How to install

Prerequisites: ensure you have Node.js and npm installed on your machine.

npm ci

Deploy the MCP server to Cloudflare Workers so it becomes reachable at your mcp URL.

npx wrangler deploy

Additional configuration and usage notes

Use the server with an MCP client by pointing to the remote URL and connecting. For example, in the Cloudflare AI Playground and similar clients, specify the MCP URL ending in /sse and establish a connection. The platform will surface a list of available tools you can invoke through natural language prompts.

If you want to connect Claude Desktop to run this server locally, you can proxy the remote MCP server using mcp-remote. This creates a local entry point that Claude Desktop can talk to, while the actual server runs remotely.

You can also configure the server in development environments like VS Code to enable Copilot-based interactions using the MCP URL.

Configure in Claude Desktop and VS Code

Claude Desktop: add a stdio MCP entry that forwards to the remote MCP URL using mcp-remote.

{
  "mcpServers": {
    "cloudflare-playwright-mcp": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://[my-mcp-url].workers.dev/sse"
      ]
    }
  }
}

Use with VS Code

Install the Playwright MCP server in VS Code to enable its use with GitHub Copilot. You can add the MCP server using the CLI integration inside VS Code.

# For VS Code
code --add-mcp '{"name":"cloudflare-playwright","type":"sse","url":"https://[my-mcp-url].workers.dev/sse"}'

Examples of interacting with the model

When you issue simple, single-action prompts, the model exchanges with the browser tools to perform the requested task. Example prompts include navigating to a page, creating a todo item, taking a screenshot, and interacting with page elements.

Available tools

browser_navigate

Navigate the browser to a given URL and update the page state for subsequent actions.

browser_type

Type text into an input field or editable element, optionally with a specified delay between keystrokes.

browser_clickt

Click a target element identified by text, selector, or label to interact with UI controls.

browser_take_screenshot

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