The Pulse
Google Home Opens Device Controls and History to AI Agents
Google’s early-access Home MCP server gives compatible AI clients a documented route to inspect smart-home structures, monitor states, run approved device actions and query event history.
AI.info Team ·
Google has opened an early-access bridge between Google Home and third-party AI agents, giving compatible assistants access to live device states, household structure, control commands, and historical event logs.
The company’s new Home Model Context Protocol server lets compatible AI clients inspect a connected home and act on it through standardized tools. Google lists Antigravity, Claude Cowork, and OpenClaw as compatible clients, while warning that connecting an agent to a real home can produce unexpected behavior.
“Connecting a real-life home to an AI agent allows that agent to control devices on your behalf.”
Google Home Developers, the documentation publisher, gives that warning in its Home MCP guide. The guide is marked “Last updated 2026-09-15 UTC,” which is an update date rather than a stated publication date. The service is separate from the company’s Home Developer MCP, which supplies coding tools with documentation covering Google Home APIs, Matter, Thread, and OpenThread.
Five tools give agents a view of the home
The consumer-facing Home MCP server exposes five main capabilities. Agents can retrieve accessible homes and structures through list_homes, then enumerate rooms, devices, traits, attributes, and command schemas with list_home_resources.
Live monitoring runs through list_home_states, which can return device connectivity and current trait states. The run_home_actions tool allows an agent to issue parameterized commands, while list_home_history queries past state changes and event logs over defined time ranges.
Google illustrates the intended uses with questions such as “How many lights do I have in my house?”, “Is my home secured?”, and “What happened while I was out?” The system can also execute commands such as turning off outside lights, subject to the client’s permission flow and Google’s restrictions.
Google requires a paid Home plan and OAuth approval
Home MCP is not an open endpoint that any assistant can call immediately. Google’s prerequisites include an active Google Home setup, an active Google Home Premium Advanced subscription, a Google Cloud project, and an MCP-compatible AI application.
Setup requires the user to enable the Home API in Google Cloud, create an OAuth consent screen, generate a web-application OAuth client, and publish the application through Google Auth Platform. The user then selects a home structure during Google’s authorization process and grants the agent access.
Google identifies separate connection paths for Antigravity, Claude Cowork, and OpenClaw. Antigravity uses a custom MCP configuration, Claude connects through its custom connectors interface, and OpenClaw uses a remote server connection with OAuth authentication.
Access can be revoked from the Google Home app or the user’s Google account page. Google also tells users to inform other household members when a shared home is connected to an agent, or to create a separate home for development and testing.
Door unlocking is off limits, automations are not ready
Google says Home MCP includes rate limits and safety protections, including a prohibition on sensitive actions such as unlocking doors. The company warns that an AI agent can still control devices and access household data, so users should review the applicable developer policies and terms before connecting a live home.
Familiar Faces data receives an additional consent step. Access requires a structure manager, a compatible Nest camera or doorbell, an active Google Home Premium subscription, and Familiar face detection enabled for the relevant camera or doorbell.
The first release also has clear limits. Creating and managing automations is not supported, although Google says it plans to add that capability in a future release. The documentation lists experimental traits as a known issue and warns that users may encounter longer-than-expected latency.
Google separates household access from developer research
Google’s wider MCP rollout includes a second server aimed at people building products around Google Home. The Home Developer MCP searches Google’s technical documentation and returns material on API methods, Matter device specifications, cloud-to-cloud requirements, and troubleshooting.
That server uses a different access model from Home MCP. Google describes Home MCP as a user-authorized OAuth service for personal agents, while Home Developer MCP offers unauthenticated trial paths and project API keys. The developer service is available through the global endpoint https://homedevelopers.googleapis.com/mcp.
The distinction gives Google two ways to place Home data in AI workflows: one connects an agent to physical devices and household history, while the other helps developers write and diagnose integrations. The first carries the greater privacy and safety burden because its tools can affect appliances, cameras, locks, and other equipment in a real home.
Home MCP puts the permission boundary in the agent client
Google’s documentation makes the MCP server a controlled interface rather than a general-purpose home API. The agent must authenticate through OAuth, the user selects the structure, and the client asks for permission before executing tools. Google’s own examples also require tool-execution approval when the assistant runs a command.
That model leaves the AI client responsible for presenting requests and results to the user while Google controls the underlying home connection. The arrangement lets assistants from different vendors work with the same Google Home environment, but it also means users must judge whether a client is interpreting a request safely before allowing it to act.
For now, the service remains early access, requires a premium subscription, omits automation management, and carries known issues around experimental traits and latency. Its initial reach is narrower than a general smart-home API, but it gives AI agents a documented route to household state, device commands, and the record of what happened at home.
Google’s Home MCP documentation provides the setup instructions, supported client configurations, tool descriptions, and current restrictions.