TypeScript MCP-MQTT SDK: Extending AI Agents from Web to IoT and Edge

(github.com)

1 points | by Kiplingbt 5 hours ago ago

1 comments

Kiplingbt 5 hours ago
We built a complete SDK ecosystem that implements Model Context Protocol (MCP) over MQTT transport (Python, C, TypeScript), but showcasing the TypeScript version today since it's easiest to integrate and test.
Why MQTT?
Existing MCP implementations are mainly HTTP-based. By introducing MQTT transport, we've expanded the boundaries of MCP AI agent applications: - Native pub/sub allows AI agents to monitor multiple data sources simultaneously - Built-in service discovery and load balancing, zero configuration needed - Lightweight protocol, better performance in unstable network environments - Cross-platform: WebSocket for browsers, TCP for Node.js
Real examples of what you can build:
Browser AI tools: Expose your browser's camera, microphone, geolocation to AI agents. Imagine ChatGPT saying "let me take a photo for you" and actually accessing your webcam through this SDK.
IoT integration: An ESP32 device running our C SDK can expose sensor data directly to AI agents. Your AI assistant can ask "what's the temperature in my workshop?" and get real-time data.
Enterprise dashboards: Any JavaScript environment can become an AI tool provider - from Electron apps to Node.js servers to browser extensions.
30-second setup:
```
    const server = new McpMqttServer({
      host: 'mqtt://localhost:1883',
      serverId: 'camera-server',
      serverName: 'browser-tools',
      name: 'Browser Camera Server',
      version: '1.0.0'
    })

    server.tool('take-photo', 'Take a photo using webcam', {
      type: 'object',
      properties: {}
    }, async () => {
      const stream = await navigator.mediaDevices.getUserMedia({video: true})
      // capture photo logic...
      return { content: [{ type: 'image', data: photoBase64 }] }
    })

    await server.start()
```
Client auto-discovers and connects:
```
    const client = new McpMqttClient({...})
    client.on('serverDiscovered', async (server) => {
      const photo = await client.callTool(server.serverId, 'take-photo', {})
      // AI now has access to your camera!
    })
```
Applications: - Smart homes: ChatGPT/Claude controlling IoT devices via MQTT - Industrial: AI quality control getting real-time sensor data - Edge AI: Lightweight AI services in resource-constrained environments - Web apps: Browser APIs exposed to AI agents (camera, location, storage) - Mobile: React Native apps becoming AI tool providers
Built by EMQX team (world's most scalable open-source MQTT broker), Apache 2.0 licensed.
Homepage: https://mqtt.ai/ GitHub: https://github.com/emqx/mcp-typescript-sdk NPM: npm install @emqx-ai/mcp-mqtt-sdk
Looking forward to HN community feedback!