Reading the tool names, you would expect `google_sheet_get` to return rows; it returns a code sample that would return rows. Only the MQTT half performs the action. That is a legitimate design for a teaching bridge — the repository is clearly aimed at IoT coursework, with a shared broker and a fixed set of endpoints — but it changes what you should connect it for.
A Node server that bundles three unrelated integrations behind one connection. The distinction that matters: the MQTT tools genuinely connect devices, publish and subscribe, while the Google Sheets and Azure AI tools return code examples showing how to call those APIs rather than calling them. The changelog records that as a deliberate change in 1.0.2.
- MQTT, which acts: `mqtt_device_create` registers a device, `mqtt_publish` and `mqtt_publish_sync` send messages with a QoS level, `mqtt_subscribe` listens on a topic, `mqtt_register_handler` attaches handler code to an action
- Google Sheets, which generates code: `google_sheet_append`, `google_sheet_get`, `google_sheet_update`, `google_sheet_delete`, `google_sheet_save` each return a working example for that operation
- Azure AI, which generates code: `azure_ai_chat` returns a sample for calling Azure AI, in either synchronous or streaming form
- Synchronous publish with a `timeout`, so a request/response exchange over MQTT is one call
- The broker and its credentials are baked into the server rather than configured
Node.js 18.0.0 or higher. Install globally straight from the repository with `npm install -g https://github.com/marty5499/mcp-api-bridge.git`, then set the command to `mcp-api-bridge` in your client config — no environment variables. Package `mcp-api-bridge`, version 1.0.2. The MQTT broker, the Sheets base URL and the Azure WebSocket endpoint are fixed in the server and point at the author's hosted services.
One command — npm install -g https://github.com/marty5499/mcp-api-bridge.git