Workers MCP Server: Extending Claude Desktop Functionality via Cloudflare Workers

Overview: What is Workers MCP Server?

The Workers MCP Server allows developers to extend Claude Desktop and other Model Context Protocol (MCP) clients by implementing custom functions within a Cloudflare Worker environment. This server leverages the new RPC syntax introduced in Cloudflare Workers, providing seamless integration between AI applications like Claude Desktop and third-party services hosted on Cloudflare or elsewhere.

🔧 Core Features & MCP Capabilities

Function Implementation

Developers can implement custom functions using TypeScript (TS) within a WorkerEntrypoint class. These functions are exposed to MCP clients, enabling rich interactions with the worker. For instance, a function like generating random numbers is straightforward:

export class ExampleWorkerMCP extends WorkerEntrypoint<Env> {
	/**
	 * Generates a super-duper random number.
	 *
	 * @return {string} A message containing a randomly generated number.
	 * */
	async getRandomNumber() {
		return `Your random number is ${Math.random()}`
	}
}

Documentation Generation

A key feature of this server is the automatic generation of documentation from TypeScript classes. This static extraction ensures that AI applications like Claude Desktop can understand and interact with the worker's capabilities accurately.

⚙️ MCP Architecture & Protocol Implementation

The Workers MCP Server comprises three core components:

1. Docs Generation: Converts TS classes into a format suitable for MCP clients.
2. Public HTTP Handler: Exposes an endpoint to handle RPC requests from MCP clients.
3. Local MCP Proxy: Facilitates communication between local and remote services.

Docs Generation

Using ts-blank-space and jsdoc-api, the server parses TypeScript files and generates a structured JSON documentation file, docs.json. This file serves as input for MCP clients to list available methods.

/**
 * Send an email with text or HTML content.
 *
 * @param {string} recipient - The destination’s email address (must be verified).
 * @param {string} subject - The subject of the email.
 * @param {string} contentType - Either "text/plain" or "text/html", indicating content type.
 * @param {string} body - The email's body, suitable for `contentType`.
 * @return {Promise<string>} A success message upon completion.
 */
async sendEmail(recipient: string, subject: string, contentType: string, body: string) {
  // Sending logic will be implemented here
}

This example JSDoc block translates into the corresponding JSON structure.

Public HTTP Handler

The lib/WorkerMCP.ts file handles incoming RPC requests via a single endpoint /rpc. It parses the request payload and delegates the method call to the appropriate worker class instance.

const workerEntrypoint = new ExampleWorkerMCP(env);
app.post('/rpc', cors(), async (req, res) => {
  try {
    const { method, args } = req.body;
    if (!workerEntrypoint[method]) {
      throw new Error(`Method ${method} not found`);
    }
    const result = await workerEntrypoint[method](...args);
    res.json(result);
  } catch (error) {
    console.error(error);
    res.status(500).json({ error: 'Internal Server Error' });
  }
});

Local MCP Proxy

The scripts/local-proxy.ts script uses the @modelcontextprotocol/sdk library to create a local proxy that simulates an MCP server. It handles tools/list requests by sending JSON data generated from the worker’s methods, and tools/call requests by making calls via /rpc.

pnpm install:claude <server-alias> <worker-url>

The above command installs a server definition for Claude Desktop that points to the local proxy.

🚀 Getting Started with Installation

1. Download Claude Desktop: Available at Claude.ai.

2. Clone this repository.

3. Install dependencies:

   pnpm install
   

4. Configure wrangler.json: If necessary, comment out sections related to unsupported bindings.

5. Deploy the worker:

   pnpm deploy:worker
   

6. Generate and upload secrets:

   npx workers-mcp secret generate && npx workers-mcp secret upload
   

7. Start integration by installing a server definition for Claude Desktop.

💡 Key Use Cases in AI Workflows

Real-Time Data Processing

Implement functions that fetch real-time data from various sources and process it using Cloudflare Workers. For example, a worker can pull stock prices from a financial API and return parsed results back to the MCP client.

Complex Prompts Generation

Develop custom prompts for AI models by combining text generation and data retrieval. This could be integrated into MC clients like Continue or Cursor to provide tailored content creation tools.

🔌 Integration with MCP Clients

MCP Client	Resources	Tools	Prompts	Status
Claude Desktop	✅	✅	✅	Full Support
Continue	✅	✅	❌	Partial Support (Tools Only)
Cursor	❌	✅	❌	Tools Only

📊 Performance & Compatibility Matrix

Mermaid Diagram for MCP Protocol Flow

graph TD
    A[AI Application] -->|MCP Client| B[MCP Protocol]
    B --> C[MCP Server]
    C --> D[Data Source/Tool]
    style A fill:#e1f5fe
    style C fill:#f3e5f5
    style D fill:#e8f5e8

Mermaid Diagram for Data Architecture

graph LR
    subgraph "Worker Environment"
        W[Worker Entrypoint]
        C1[Cloudflare API Calls]
    end
    subgraph "MCP Client"
        P[MCP Protocol]
        S[Server (Local Proxy)]
    end
    A[Data Source/Tool]-->B[P]
    B --> S
    S --> W
    W --> C1

🛠️ Advanced Configuration & Security

API Key and Secret Storage

{
  "mcpServers": {
    "[server-name]": {
      "command": "npx @modelcontextprotocol/server-[name]",
      "args": ["--api-key", "your-api-key"],
      "env": {
        "API_KEY": "your-api-key",
        "SECURE_SECRET": "your-secure-secret"
      }
    }
  }
}

Ensure all sensitive information is securely stored and managed.

❓ Frequently Asked Questions (FAQ)

1. Does this work with all MCP clients?

   • The server supports a wide range of MCP clients, including Claude Desktop and Continue, but some clients may have partial support due to limitations in their functionality.

2. How do I troubleshoot connection issues?

   • Check the status codes returned by the worker and ensure that the environment variables are correctly set.

3. Can I add more tools without re-deploying the codebase?

   • Yes, you can extend the worker with additional methods by modifying the TypeScript source files and redeploying them to Cloudflare.

4. What if my data sources change frequently?

   • You can dynamically update the worker logic via serverless functions that call newer APIs or datasets without reuploading the entire codebase.

5. How do I ensure security of my API keys?

   • Store all sensitive data in environment variables and avoid hardcoding them in source files to prevent accidental disclosure.

👨‍💻 Development & Contribution Guidelines

Getting Started

1. Clone this repository.
2. Install dependencies: pnpm install.
3. Write methods within the WorkerEntrypoint class.
4. Deploy changes with pnpm deploy:worker.

Collaboration

• Raise issues or send pull requests to improve this server.
• Engage with the community for real-world use cases and feedback.

🌐 MCP Ecosystem & Resources

Explore more about MCP and its ecosystem at ModelContextProtocol.org.

For further reading and detailed documentation, visit:

• Cloudflare Workers Documentation
• MCP Protocol Specifications

This comprehensive guide highlights the capabilities and integration points of the Workers MCP Server, positioning it as an essential tool for developers looking to enhance AI applications through standardized protocols.