Since September 16, Google Home has offered an MCP server that lets third-party AI agents such as Claude, ChatGPT and OpenClaw read the state of a home and control its devices. Early access is limited to US subscribers of Google Home Premium Advanced, which costs $20 a month.
Key Takeaways
- Home MCP gives agents five tools covering homes, devices, live states, actions and event history.
- Sensitive actions such as unlocking doors stay off limits.
- Setup runs through a Google Cloud project, which points first at technical users.
Have an AI Sum Up This Article
ChatGPTWhat Google Home Now Hands Over to AI Agents
MCP, the open standard that lets a model call tools and services, has reached Google’s smart home. Google’s developer documentation for the Home MCP server describes it as a bridge between a household setup and AI assistants.
That matters because MCP is built for exactly this job: letting AI models and chatbots talk to apps and services and fetch data from them in a common format. Without a shared protocol, every assistant maker would have to build and maintain its own Google Home integration, one agent at a time.
The server exposes five tools. An agent can list the homes it can reach, enumerate devices and their command schemas, read live device states, run actions and query how states changed over a time range. In the documentation those tools carry explicit names: list_homes, list_home_resources, list_home_states, run_home_actions and list_home_history.
In plain language, a user can ask for camera summaries, keep an eye on activity around the house, adjust a device or build a custom dashboard. Google points to Nest doorbells and thermostats, plus Works with Google Home devices built on Matter, such as smart bulbs.
The agents named so far are Claude, ChatGPT, Hermes, OpenClaw and Google Antigravity. In principle, any agent that can call MCP tools can connect once the account holder grants permission.
Google has already tested the idea of an agent acting on a user’s behalf with Remy, the personal agent project pitched as running around the clock. Home MCP turns the house into concrete ground for that kind of agent, without requiring Google’s own.
The history tool is the most novel piece. Querying state changes over a period lets an agent work out when a thermostat was adjusted or when a doorbell rang, then line those events up with camera summaries.
Access Stays Narrow and Guarded
Getting in takes some effort. Users need a Google Home Premium Advanced subscription at $20 a month, a tier that already bundles longer event video history, descriptive notifications, detailed alerts, video search and daily summaries.
Setup means creating a Google Cloud project, enabling Home MCP and the Home API, generating OAuth credentials, then passing the configuration to the chosen agent, which asks the user to sign in and grant permissions. Google’s Home Developer Center carries the step-by-step guide.
Those steps filter the audience. A household that simply wants to talk to its lights will not open a Cloud console, so the first wave of Google Home agents will come from developers, tinkerers and power users willing to manage credentials. Consumer-grade setup would be the signal that Google wants this in every home.
Guardrails come built in. The server enforces rate limits and blocks sensitive actions, with door unlocking given as the example. No agent gets in without the account holder signing in and granting permissions during setup, which keeps the subscriber in charge of which outside assistant touches the house and its connected devices.The documentation also warns that experimental traits may misbehave and that latency is still being optimized.
The rollout reaches US subscribers over the coming weeks, and Google has not confirmed when other countries will follow. Feedback is being collected through the Smart Home for Developers community.
Tying the feature to the top Google Home tier also gives that plan a new selling point. Subscribers were paying for video history and summaries, and they now get the right to plug an outside agent into the same data, which makes the $20 monthly fee easier to justify for heavy users.
More articles on Horizon
- Claude Cowork and Chat Merge Into One Claude App
- Gemini 3.8 Live Undercuts OpenAI on Voice AI Pricing
- Siri AI on iPhone Leaves EU Users Waiting
Google Home Turns Into Plumbing for Every Assistant
The decision stands out for a company pushing Gemini everywhere. Google retired its original Assistant on September 4 in favor of Gemini, yet it is now opening the smart home to rival assistants.
For users, the appeal is choice. Someone who already relies on Claude or ChatGPT at work can hand that same agent the thermostat or the camera feed. Google already supports MCP across its Cloud platforms, developer tools and Google Workspace, and the home now joins that list.
The custom dashboard use case also shifts who designs the interface. Instead of waiting for Google to add a view to its app, a subscriber can ask an agent to assemble one around the devices and events that matter in that particular house.
For competitors, the move frames Google Home as infrastructure more than as an assistant in its own right. It echoes the desktop race Horizon covered with Gemini Spark taking on Claude and Copilot on the Mac, only reversed, since Google is now letting a rival’s agent act inside its own customers’ homes.
Matter device makers stand to gain as well. Their products become controllable by any agent routed through Google Home, which strengthens Google’s position at a moment when assistants keep multiplying.
Amazon is taking the opposite route with its own assistant. On the same day, it launched Alexa+ in India with Hindi support, pitching a single in-house assistant that handles multi-step tasks such as ordering groceries and switching smart home devices on or off. Two strategies now coexist in the connected home, one built around a house assistant and one built around an open door for outside agents.
Trust is the open question for the weeks ahead. An agent that reads camera history touches intimate data, and Google will need to spell out how far its guardrails reach beyond front doors. Early feedback from developers in the access program should offer the first clues, well before any wider launch outside the United States.
Follow the story on Horizon.
