Google is opening up Google Home to outside AI agents with a new Model Context Protocol integration. The update lets agents that support MCP, including Claude, Hermes, OpenClaw, ChatGPT, and Google Antigravity, work with devices in the Google Home ecosystem and access event history.
That means a user can ask an agent to review camera summaries, monitor smart home activity, control connected devices, or build a custom dashboard. Google says the system also supports devices like Google Nest doorbells and thermostats, plus 'Works with Google Home' devices, including Matter light bulbs.
Getting it running is not exactly plug-and-play. Users need to create a Google Cloud project, configure it for Home MCP, and then hand the configuration details to their chosen agent. The agent then asks the user to sign in and grant permissions. Google says a setup guide will also be available in the Google Home Developer Center.
The rollout starts today and will continue over the coming weeks, but only for Google Home Premium Advanced subscribers in the U.S. That tier costs $20 per month and includes longer event-based video history, descriptive notifications, detailed alerts, tools to search video history, and daily summaries. Google would not say whether MCP will expand to other subscription tiers or markets.
Google says the MCP server can be used to let an AI agent securely work with smart home devices, but it is also drawing a line around what those agents can do. The company says Home MCP enforces rate limits and safety protections, and says it will not allow an agent to unlock doors. Taylor Lehman, group product manager at Google Home & Nest, also warned that connecting an agent to Home MCP can lead to 'unexpected or even undesired behavior.'
That warning matters because the system reaches into real-world infrastructure: smart locks, thermostats, HVAC systems, and appliances. In other words, this is not just another chatbot feature bolted onto the side of a walled garden. It is an attempt to let third-party agents operate on the underlying data and control layer of the home.
Google has been pushing the platform toward that role for a while. It launched API access to its smart home in 2024 and has been building Gemini for Home as a fuller AI layer on top. Home MCP adds another layer of control, which also fits Google’s broader developer play: provide the infrastructure, then let other companies build on top.
For users, the pitch is obvious. Natural language is easier than wrestling with clunky menus when you want to inspect logs, debug automations, or figure out why a device misbehaved. But the same capability that makes a smart home more useful also makes it more exposed. Once an agent can reason across your home’s history and act on your behalf, the platform stops being a dashboard and starts becoming an operating layer.
Nobody asked for a home computer that can think across your camera footage, device history, and automations. But Google is shipping one anyway. The only question now is whether the convenience is worth the permission prompts, the rate limits, and the security trade-offs baked into the changelog.
