Google is giving smart-home owners a rather interesting new option: letting an AI agent of their choice run the house. The company has introduced Google Home MCP, a new Model Context Protocol integration that allows compatible AI agents such as Claude, Google Antigravity, Hermes, and OpenClaw to access and control devices connected to Google Home. It can also access device history, meaning the AI isn’t limited to simply switching a light on or off.
Your AI assistant just got access to the house
The idea is surprisingly simple. Instead of talking exclusively to Gemini through the Google Home app, users can connect another MCP-compatible AI agent and let it interact with their smart home.
That opens up some genuinely useful possibilities. An agent could check whether the home is secured, turn off a collection of lights, look through camera and device history, or analyze patterns across multiple devices. Google even gives examples such as asking what happened while someone was away or how many loads of laundry were done during the week.
The agent can also create custom dashboards and deliver information through Google Home speakers. So instead of remembering which app controls which gadget, users can increasingly just ask an AI to figure it out.
But maybe don’t give the robot all the keys just yet
Home MCP is currently in early access for Google Home Premium Advanced subscribers in the US, with access rolling out over the coming weeks. The plan costs $20 per month or $200 per year, while setup currently requires creating a Google Cloud project and configuring Home MCP.
Google has also built in some safety limits, including blocking AI agents from unlocking connected doors. Still, the company cautions that the experience can vary by agent, and, as Lehman puts it, “connecting it to Home MCP can result in unexpected or even undesired behavior.” He advises users to go through Google’s developer policies and terms of service before handing an AI agent access to their smart home.
At the end of the day, Google is turning Home into a platform that different AI agents can plug into, rather than tying smart-home control exclusively to Gemini. The bigger idea is that smart homes could move from simply following commands to understanding what’s happening, reasoning about it, and taking action. The only sensible rule for now? Give the AI the keys one permission at a time.
