Google has opened an MCP server for Google Home, giving AI agents access to connected devices and to the event history behind them. Through it, an agent can pull camera summaries, track activity in the house, control devices and build custom smart-home dashboards, all driven by a plain-language description of what the user wants. Access begins today and widens over the coming weeks — but only for Google Home Premium Advanced subscribers in the United States, the most expensive tier at $20 a month.
Setting it up is not a consumer flow. The user first creates a Google Cloud project and configures it for Home MCP. They then hand the MCP configuration parameters to their chosen AI agent and ask it to run the setup, after which the agent prompts them to sign in and grant the necessary permissions. A step-by-step guide is due in the Google Home Developer Center.
Once connected, the server covers the whole Google Home ecosystem: Nest doorbells and thermostats, plus anything carrying "Works with Google Home" or Matter support, down to light bulbs.
Google has been fitting MCP endpoints across its products for a while — Google Cloud, its data platforms, developer tools and Workspace all have them. Home support is aimed, in the company's framing, primarily at people already experimenting with AI agents for everyday tasks.
The tier requirement is the most revealing part of the announcement. Premium Advanced is Google's video-heavy subscription: longer event-linked video history, descriptive notifications, detailed alerts, search across video history, daily summaries. Those are the features that produce the event data an agent would want to query — so gating MCP behind that tier has a technical logic. It also means the feature launches to the smallest, highest-paying slice of the user base, and Google said nothing about whether MCP will reach other tiers, other countries, or a broader audience at all, or when.
What strikes me about the setup instructions is who they describe. Creating a Google Cloud project, configuring it for Home MCP and pasting configuration parameters into an agent is developer work. Paying $20 a month for camera summaries and daily digests is consumer behavior. The overlap between those two groups is small, and Google is collecting early-access feedback through its Smart Home for Developers community, which suggests the company knows exactly which of the two it expects to show up. This reads less like a product launch than a structured beta wearing a subscription tier as an access-control mechanism.
The choice of protocol is the part worth taking seriously. Google is not describing this as a feature of its own assistant; the instructions say to hand the configuration to the agent the user has chosen. A home platform deliberately exposed through a protocol any competitor's agent can speak is a real concession about where the value sits — in the devices and the event history, not in whose assistant reads them.
Notably absent from the announcement is any account of what an agent is allowed to get wrong. The permissions step is described only as a prompt during setup; there is nothing on scope limits, on whether an agent can act on doorbells and thermostats as freely as it reads camera summaries, or on what a misinterpreted instruction costs. In a chat window a bad tool call produces a wrong answer. In a house it changes the state of something physical, and the user finds out later.
That gap is the actual story of agents reaching physical devices, and it will not stay theoretical for long. Google has chosen to find out on its highest-paying US subscribers first, which is the sensible way to run the experiment — and also the reason the rollout schedule past those subscribers is the one thing the company declined to describe.