Google is opening up a new way for AI agents to talk to your smart home, but it is not a simple toggle in the Google Home app. On September 16 the company opened Early Access to Home MCP, a Model Context Protocol server that lets outside AI agents read and control devices tied to a person’s Google Home setup. There is also a separate Home Developer MCP aimed at engineers and coding tools. Home MCP is the one that creates a path for third‑party AI systems to understand a home’s structure, watch what devices are doing, and send commands back in a controlled way.
What Home MCP actually does for AI agents
Home MCP is designed as a bridge between a person’s Google Home and an outside AI agent, whether that agent is Google’s own or another company’s. Once connected, that agent can perform what Google calls structure discovery, resource discovery, state monitoring, device control, and historical analysis. In regular language, the agent can list out the homes it is allowed to reach, see the rooms and the devices inside each one, watch their current status in real time, issue control commands, and look back over earlier states and time‑ordered logs.
Google describes the server as a proxy that sits between smart home infrastructure and AI applications, exposing standard MCP tools and RPC interfaces that let an AI client query live home telemetry and issue commands. Because the server uses the Model Context Protocol, it can work with any MCP‑compatible AI client or application, as long as the user sets it up correctly. Google’s own product page says the design is meant to meet its security and privacy standards so agents cannot take sensitive actions on your behalf.
Two different MCP servers: Home and Home Developer
Google is not shipping just one MCP endpoint. Home MCP is the part meant for everyday users who already run their homes through Google Home and want to connect a personal AI agent to their own setup. That is the server that can watch device states, read past events, and send commands inside a private home.
Home Developer MCP is a separate server aimed at software and firmware engineers. Instead of connecting to a physical home, it lets those tools search the Google Home Developer Center, OpenThread, and the Matter and Thread specifications, feeding that material into coding agents such as Claude Code, Cursor and GitHub Copilot. That developer‑side server is meant to help those tools reason about the Google Home ecosystem and write code against the documented interfaces, while Home MCP focuses on real devices in real homes.
Safety limits and how Google says it protects your home
Google states that Home MCP is not a free‑for‑all pipe into your devices. The company says the server enforces rate limits and safety protections on what agents can do and how often they can do it. As its own example, Google says sensitive actions like unlocking doors are prohibited through this channel.
Google does not publish a full list of restricted actions, and unlocking doors is the only one it names outright as prohibited. The documentation does add a second concrete guardrail: reaching familiar faces data takes a separate, explicit consent, and only a manager of the home structure can grant it. Google’s own warning is blunter than the safeguards suggest: connecting a real home to an agent lets that agent control devices on your behalf, and Google says that depending on the agent, this can result in unexpected or even undesired behavior. Google also asks anyone connecting a home shared with other household members to tell them that the agent can control devices and access home data, and notes that an agent’s access can be revoked at any time from the Google Home app or your My Accounts page.
What you need and how setup works
This is not a simple consumer feature where you tap a card in the Google Home app and everything starts working. To use Home MCP, Google says you need an active smart home setup in Google Home with connected devices and an active Google Home Premium Advanced subscription specifically, not just the standard Premium plan. Google said at launch that access would roll out over the following weeks to Premium Advanced subscribers in the US, so an otherwise eligible account may not have it yet. You also need access to a Google Cloud project and an MCP‑compatible AI client or application such as Google Antigravity, Claude Cowork or OpenClaw.
The setup path runs through Google Cloud, not only through the Home app. Google says you start by creating a Google Cloud project, then obtaining access approval and enabling the Home API in the Google Cloud console for that project. After that, you configure OAuth consent, create OAuth client credentials with redirect URIs, and publish the app under the Google Auth Platform’s Audience settings, so that your chosen MCP‑compatible AI client can request access and receive tokens in a standard OAuth flow. Only once all of that is in place can your AI agent connect to Home MCP and begin discovering your home structure, reading device states, and issuing allowed commands under your authorization.
How an AI agent can see and use your smart home data
Once a user has gone through the cloud setup and linked an MCP‑compatible client, Google says the agent gains a fairly detailed view of the home. Structure discovery lets the agent retrieve the homes and structures it can reach. Resource discovery then pulls the devices, area layouts, traits, attributes and command schemas tied to that structure.
State monitoring and historical analysis then give the agent time‑based awareness. It can watch the current state of devices as they change and look back over past states and chronological event logs. Device control closes the loop, allowing the agent to issue commands to devices inside the policies and limits Google enforces. Together, these pieces are meant to let an external AI agent act with context about how a home is arranged and how it behaves over time, while still passing through Home MCP’s rate limits and safety filters. One gap is worth knowing before you invest the setup time: Google says creating and managing automations through Home MCP is not supported at this time, and lists automation support as coming in a future release.
View the Google Home MCP documentation.
