Demonstration of AI Tool Integration via Workers MCP

This repository showcases a minimal implementation project utilizing the Cloudflare Workers MCP framework. The resulting utilities are designed to function seamlessly with advanced conversational agents such as Claude, Cursor, and any other platform adhering to the MCP specification.

Understanding MCP

MCP, or the Model Context Protocol, is a standardized specification, openly released by Anthropic. Its primary function is to establish robust communication channels between sophisticated AI models and external data repositories, organizational tools, or development environments. The core objective is to furnish cutting-edge models with requisite context to yield superior, contextually relevant responses.

This specific project leverages Cloudflare Workers to host these customized tools, making them callable and usable by AI assistants, thereby dramatically expanding the operational reach of the AI.

Key Features

• Custom-built AI utilities hosted on the Cloudflare Workers runtime.
• Native, friction-free integration with clients like Claude, Cursor, and broader MCP consumers.
• Capacity to augment AI functionality through proprietary, on-demand serverless functions.

Prerequisites for Setup

To successfully build and operate this solution, you must have the following installed:

• Node.js (Version 16 or newer is required)
• The pnpm package manager
• An active Cloudflare Account
• The Wrangler CLI
• An MCP-compatible agent (e.g., Claude Desktop, Cursor)

Implementation Guide

1. Obtain the Source Code

Clone this repository locally:

bash git clone https://github.com/akazwz/workers-mcp-demo.git cd workers-mcp-demo

2. Dependency Installation

Install necessary packages:

bash pnpm install

3. Cloudflare Worker Configuration

Ensure you are authenticated via Wrangler:

bash pnpx wrangler login

Execute the setup utility to initialize MCP support for your Worker:

bash npx workers-mcp setup

This command handles the necessary boilerplate configuration for MCP compliance.

Note for Complimentary Tier Users

If you are utilizing the Cloudflare Workers Free Plan, you are required to manually adjust wrangler.jsonc by either removing or commenting out the following configuration block:

c "browser": { "binding": "BROWSER" }

This step is necessary because the Browser Rendering API is exclusively reserved for paid Cloudflare Worker subscriptions. Removal will disable image capture (screenshotURL) and extraction (extractImagesFromURL) features, though all other implemented utilities will remain fully operational.

4. Deployment

Deploy your configured Worker to the Cloudflare network:

bash pnpm run deploy

5. Client Integration (MCP Agent Setup)

For Claude Desktop Users:

Execute this command to register your new MCP server within Claude:

bash pnpx workers-mcp install:claude

For Cursor Users:

In Cursor's MCP configuration settings, incorporate the following definition:

{ "type": "command", "command": "/path/to/workers-mcp run workers-mcp-demo https://your-worker-url.workers.dev /path/to/workers-mcp-demo" }

Usage Instructions

Once setup is complete, you can invoke your custom utilities directly through the AI assistant interface. For instance, in Claude Desktop, the invocation might look like this:

@workers-mcp-demo getRandomNumber

Available Functionality

This demonstration package exposes the following callable functions:

• getRandomNumber: Generates a pseudo-random numeric value.
• generateImage: Synthesizes an image based on a provided textual prompt.
• screenshotURL: Captures a visual snapshot of a specified web address (requires paid account).
• extractImagesFromURL: Scans a webpage and pulls out embedded graphical assets (requires paid account).

Customization

To integrate your own logic, modify the TypeScript source file located at src/index.ts by appending new methods to the Worker class. After modification, re-deploy the service using:

bash pnpm run deploy

Troubleshooting Common Issues

• Agent Not Discovering Tool: If the AI assistant fails to recognize the new tool, try restarting the assistant application.
• Changes Not Reflected: If you modify a function's signature or name, a restart of the AI agent is usually required for the changes to propagate.
• Deployment Errors: Inspect the Cloudflare Workers logs for any runtime or deployment-related exceptions.
• Free Tier Browser Error: If you encounter errors related to the Browser Rendering API while on a free plan, consult the configuration notes in Step 3 regarding wrangler.jsonc adjustments.

Licensing

This project is distributed under the terms of the MIT License.

Acknowledgements

We extend gratitude to the following foundational technologies:

• Cloudflare Workers MCP for the core integration plumbing.
• Anthropic Model Context Protocol for establishing the open standard.
• Cloudflare Workers for providing the global serverless compute platform.