Google Home MCP has entered Early Access in the United States, giving compatible third-party AI clients a way to connect to a Google Home environment and work with its devices, current states and event history. The rollout began on September 16, 2026, in U.S. English, with access rolling out over the following weeks.
This is not a new smart speaker or a replacement for Gemini in the Google Home app and Nest speakers. It is a developer-oriented connection that acts as a bridge between Google Home and AI applications such as Google Antigravity, Claude Cowork and OpenClaw, which Google lists as compatible clients.
What Google Home MCP lets an AI agent access
The Model Context Protocol, or MCP, is a standardized way for an AI application to discover and use external tools and data. In Google Home, the connection exposes five practical capability areas:
| Capability | What the authorized client can do |
| Home discovery | Find the Google Home environments available to the account |
| Resource discovery | Inspect devices, areas, traits, attributes and command schemas |
| State monitoring | Read connectivity information and current device states |
| Device control | Send parameterized commands to supported target devices |
| Historical analysis | Review past state changes and event logs over a specified time range |
That combination gives an agent more context than a simple voice command. It can, for example, work across the structure of a home, check current device states, examine past events or issue an action to a compatible device. Google’s documented ecosystem examples include Google Nest cameras and doorbells, Nest Thermostats and Matter-compatible lights.
The connection also opens the door to more involved tasks, such as analyzing activity across rooms, summarizing camera history or building a custom smart-home dashboard. The exact functions available can vary by client; compatibility does not mean every client exposes the same tools.
The setup is an Early Access developer connection
To use Home MCP in the United States, you need an active Google Home Premium Advanced subscription, a Google Cloud project with the Home API enabled, an OAuth consent screen, OAuth client credentials and an MCP-compatible AI client. User consent is part of the OAuth 2.0 authorization flow.
Google Home Premium Advanced is listed at $20 per month or $200 per year in the United States. The subscription is one part of the setup; connecting the service also involves Google Cloud configuration and permissions for the selected client.
This makes Home MCP substantially different from opening the Google Home app and asking Gemini to control a device. The standard Google Home experience remains the familiar consumer interface, while Home MCP is aimed at connecting an external agent to the home’s tools and data.
The limits matter in a physical home
Giving an AI agent access to lights, cameras, thermostats and other connected devices is useful only if the boundaries are clear. Google documents rate limits and safety protections, including restrictions on sensitive actions such as unlocking doors.
Google also warns that an agent’s behavior can be unexpected or undesirable. Access can be revoked through the Google Home app or the user’s My Accounts page, and Google recommends reviewing the relevant developer policies and terms of service before granting permissions. The documentation also warns that experimental traits may not work as expected and that some operations may have higher-than-expected latency.
Home MCP does not currently support creating or managing automations. Google says that capability is planned for a future release, so the Early Access feature currently focuses on discovering homes and resources, reading states and history, and running supported device actions.
What the September rollout means
The September 16 rollout puts Google Home on a path toward broader agent access, but its first U.S. implementation remains a subscription-linked developer connection rather than a one-tap consumer setting. For smart-home owners who already use an MCP-compatible client, the important change is the new access layer: an authorized agent can work with Google Home’s structure, device data, controls and history from outside the standard Home interface.
