Cloudflare Workers

🚀 workers-mcp

Connect with Claude Desktop via Cloudflare Worker!

⚠️ Important Note

You should start from here instead of building a remote MCP server.

You can use mcp-remote to connect to the remote MCP server of your account from Claude Desktop, Cursor, and other clients here.

🚀 Quick Start

What is workers-mcp?

This package provides CLI tools and the logic to run in a Worker, enabling you to connect Claude Desktop (or other MCP clients) to the Cloudflare Worker in your account. This can be customized to meet your needs. It transforms TypeScript methods like the following code into MCP tools through a build step:

export class ExampleWorkerMCP extends WorkerEntrypoint<Env> {
  /**
   * Generate a random number. Since this number must pass through Cloudflare's nearest POP to be calculated, this number is extra random... something about lava lamps.
   *
   * @return {string} A message containing a super random number
   * */
  async getRandomNumber() {
    return `Your random number is ${Math.random()}`
  }
  
  // ...and so on
}

...into MCP tools for a local Node.js server to expose these tools to MCP clients. The Node.js server acts as a proxy, handling local stdio transportation and invoking the relevant methods in the Worker running on Cloudflare. In this way, you can re - offer any function or API in your application, or any service in the Cloudflare Developer Platform, to LLMs, Claude Desktop, or other MCP clients.

Here, we notice that the random number generation is actually not random because all calls return the same value. This is a known issue.

📦 Installation

To get started, follow these steps:

1. Generate a new Worker project using npx create-cloudflare@latest my-new-worker.
2. Navigate to the project directory and run npm install workers-mcp to install the required dependencies.
3. Run npx workers-mcp setup to set up the MCP configuration. If you encounter issues, refer to the help information for debugging.

💻 Usage Examples

Basic Usage

Every time you modify the Worker code, simply run npm run deploy to update Claude's metadata and the actual Worker instance.

Advanced Usage

For other clients like Cursor, you need to combine the command and parameters into a string and specify the type as command. Ensure that the path and parameters are correct to avoid connection issues.

📚 Documentation

Example Projects

• examples/01-hello-world demonstrates the initial state after installation.
• examples/02-image-generation uses Workers AI to run an image generation model, and Claude can adjust the prompt based on the results.

Notes

The documentation recommends using a remote MCP server instead of a local proxy, which may perform better in terms of stability and performance. Understanding the specific steps and advantages of remote configuration can help you optimize your setup.

Overall, workers-mcp is a powerful toolchain that can directly provide the capabilities of Cloudflare Worker to AI assistants like Claude. You may encounter path errors, permission issues, or connection timeouts during actual operations. Refer to the troubleshooting section in the documentation to solve these problems.