Linkly AI MCP Remote Worker

English | 中文 | 日本語 | 한국어 | Español | Français | Deutsch

Expose your Linkly AI Desktop local MCP (Model Context Protocol) server to the internet via Cloudflare Workers.

This worker acts as a reverse proxy, allowing remote MCP clients (like ChatGPT, Claude, etc.) to access your local MCP server running on your desktop through a secure WebSocket tunnel.

Architecture

Remote MCP Client (ChatGPT/Claude...)
    ↓ HTTPS POST /mcp
Your Cloudflare Worker
    ↓ WebSocket (Hibernatable)
Durable Object (Tunnel Manager)
    ↑ WebSocket (Long-lived connection)
Linkly AI Desktop App
    ↓ HTTP
Local MCP Server (127.0.0.1:60606/mcp)

Deploy

One-Click Deploy

Manual Deploy

1. Clone this repository:

   git clone https://github.com/LinklyAI/linkly-ai-mcp-remote-worker.git
   cd linkly-ai-mcp-remote-worker

2. Install dependencies:

   pnpm install

3. Login to Cloudflare:

   npx wrangler login

4. Deploy the worker:

   pnpm run deploy

5. Note down your worker URL (e.g., https://linkly-ai-mcp-remote.<your-account>.workers.dev)

Usage

Configure Linkly AI Desktop

1. Open Linkly AI Desktop
2. Go to Settings → MCP tab
3. Enter your worker URL (without https:// prefix)
4. Click Test to verify the connection
5. Click Save to save the configuration
6. Enable the tunnel switch on the MCP page

Configure MCP Clients

Once the tunnel is connected, configure your MCP client to use the remote endpoint:

Claude Desktop (claude_desktop_config.json):

{
	"mcpServers": {
		"linkly-ai": {
			"url": "https://linkly-ai-mcp-remote.<your-account>.workers.dev/mcp"
		}
	}
}

Cursor (.cursor/mcp.json):

{
	"mcpServers": {
		"linkly-ai": {
			"url": "https://linkly-ai-mcp-remote.<your-account>.workers.dev/mcp"
		}
	}
}

API Endpoints

Endpoint	Method	Description
/	GET	Worker info and available endpoints
/health	GET	Health check (shows tunnel status)
/tunnel	GET	WebSocket endpoint for desktop tunnel
/mcp	POST	MCP endpoint for remote clients

Development

Local Development

pnpm run dev

This starts a local development server at http://localhost:8787.

Generate Types

pnpm run cf-typegen

Run Tests

pnpm test

How It Works

1. Desktop Connection: Your Linkly AI Desktop app connects to the worker via WebSocket at /tunnel
2. Hibernation: The Durable Object can hibernate during idle periods, reducing costs while keeping the WebSocket alive
3. MCP Proxy: When a remote MCP client sends a request to /mcp, the worker forwards it through the WebSocket to your desktop
4. Local Forwarding: Your desktop app receives the request and forwards it to your local MCP server
5. Response Path: The response travels back through the same path

Security Considerations

• No Authentication: The current version has no authentication. Your worker URL acts as a simple access control.
• Self-Deployed: Each user deploys their own worker instance, providing isolation.
• HTTPS Only: All communications use HTTPS/WSS encryption.

For production use, consider adding authentication via Cloudflare Access or API keys.

Cost

This worker uses Cloudflare Durable Objects with WebSocket Hibernation, which is cost-effective:

• Workers: Free tier includes 100,000 requests/day
• Durable Objects: Pay-as-you-go, but hibernation minimizes active time
• WebSocket: Auto ping/pong doesn't wake the Durable Object

Typical personal usage should stay within free tier limits.

License

MIT License - see LICENSE for details.

Related Projects

• Linkly AI Desktop - AI-powered knowledge management desktop app
• Model Context Protocol - Open protocol for AI model context