Google opens Home devices to AI agents through new MCP server

Google has begun rolling out early access to a server built on the Model Context Protocol, a standard connecting AI agents to outside tools, that lets AI agents control devices in the Google Home ecosystem. The company announced the rollout on Wednesday, according to TechCrunch.

The MCP server lets any AI agent that supports the protocol (including Claude, ChatGPT, Hermes, OpenClaw, and Google’s own Antigravity) work with a household’s connected smart home devices and pull up their event history. People will be able to type or speak plain-language requests to review camera summaries, check on smart home activity, control individual devices, or build custom dashboards for their home setup.

Setting up the connection takes a few steps. Users must create a project in Google Cloud, Google’s cloud computing platform, and configure it to work with Home MCP, then hand the resulting configuration details to whichever AI agent they want to use and ask it to complete the setup. The agent prompts the user to sign in and grant permissions before it can act on the household’s devices, and Google says it will publish a setup guide through its Home Developer Center. The system covers any device already in the Google Home ecosystem, including Nest doorbells and thermostats, as well as third-party products that carry the “Works with Google Home” or Matter certification, such as smart light bulbs.

Limited to paying subscribers, for now

Access to the MCP server is not open to every Google Home user. Google is starting with subscribers to Google Home Premium Advanced, the service’s $20-per-month tier, in the United States only. That plan already includes longer video history tied to detected events, more descriptive notifications, tools for searching past footage, and daily summaries of home activity. Google has not said whether or when it will extend MCP access to cheaper subscription tiers or to markets outside the US.

This is not Google’s first push into the protocol. The company has already added MCP support to Google Cloud and its data platforms, its developer tools, and Google Workspace, but bringing it to Google Home marks a more consumer-facing step, aimed at people already experimenting with AI agents to handle daily tasks rather than developers building on Google’s infrastructure. Google is treating the release as a trial run, collecting feedback from early users through its Smart Home for Developers Community, and has left open how the rollout will expand. What happens next will depend on that feedback: whether Google widens access beyond Premium Advanced subscribers, adds support for more device types, or keeps the feature limited while it works out how agents should be allowed to act on a home’s devices.