On Tuesday, Google began rolling out early access to a Model Context Protocol server for Google Home, opening its smart-home ecosystem to compatible AI agents. The move means tools including Claude, ChatGPT, Hermes, OpenClaw and Google Antigravity can, with user permission, examine event histories and interact with connected devices. But the first wave is tightly gated: it is available in the U.S. to Google Home Premium Advanced subscribers, the $20-a-month tier, with no broader timetable announced.¹

Google’s case is that MCP makes the smart home less like a collection of isolated gadgets and more like an environment an agent can understand. Taylor Lehman, a Google Home and Nest product manager, said the integration lets supported agents “securely work with all of the devices and event history in your Google Home ecosystem.”² In practice, Google envisions questions that stretch beyond switching off lights: an agent could review cross-camera activity, calculate how long lights were left on, track laundry loads, create a custom dashboard or send a spoken update through a Home speaker once a task is finished.

The setup still demands some technical work. Users must create a Google Cloud project, configure Home MCP, pass configuration details to their chosen agent, then sign in and approve permissions. That makes the launch an early-adopter experiment as much as a consumer feature; Google is collecting feedback through its developer community.¹

For proponents, the attraction is clear: natural-language control and analysis could turn years of device data into something useful. Yet the enthusiasm arrives amid a darker public argument about AI’s expanding role. One observer said they would be “quite excited” by the integration were the week not dominated by warnings that “AI will kill us all,” framing the promise of an agentic home against anxieties that reach far beyond thermostats and doorbells.³