Google Home MCP: Requirements, AI Agent Access and Safety Limits
Google has opened early access to a Google Home Model Context Protocol server that lets compatible AI agents inspect smart-home devices, monitor live state, execute supported device actions and query historical events. The rollout adds a standard MCP interface between Google Home and agent clients including Google Antigravity, Claude Cowork and OpenClaw.
Access currently requires an active Google Home Premium Advanced subscription, a Google Cloud project and an MCP-compatible client. Google’s developer documentation marks Home MCP as Early Access, while current launch coverage places the initial rollout with Premium Advanced subscribers in the United States.
Home MCP exposes five core tool areas: home and structure discovery, device/resource discovery, real-time state monitoring, device control and historical event analysis. Google also applies rate limits and blocks sensitive actions such as unlocking doors.
Google Home MCP requirements at a glance
| Requirement | Current early-access requirement |
|---|---|
| Google Home | Active home with connected devices |
| Subscription | Google Home Premium Advanced |
| Cloud setup | Google Cloud project with Home API enabled |
| Authentication | OAuth 2.0 |
| Client | MCP-compatible AI client |
| Google examples | Antigravity, Claude Cowork, OpenClaw |
| Server | https://home.googleapis.com/mcp |
| Device control | Supported through MCP tools |
| Historical events | Supported |
| Sensitive actions | Restricted; Google cites door unlocking as blocked |
| Automations | Creation and management are currently unsupported |
Google’s documentation was last updated September 15, and the early-access rollout was announced September 16, 2026. TechCrunch reports that access is rolling out over the coming weeks to Google Home Premium Advanced subscribers in the U.S.; that tier is currently priced at $20 per month.
What an AI agent can access
The server acts as a proxy between an MCP client and Google Home infrastructure. Google publishes five core tool groups:
| MCP capability | What it exposes |
|---|---|
list_homes |
Homes and structures available to the user |
list_home_resources |
Devices, areas, traits, attributes and command schemas |
list_home_states |
Current connectivity and device trait state |
run_home_actions |
Parameterized control actions on supported devices |
list_home_history |
Past state changes and event logs over selected time ranges |
This creates a wider agent workflow than a single voice command. An agent can discover the available device topology, inspect current state, correlate historical events and then execute an allowed action. Google’s own verification examples include asking how many lights are present, checking whether a home is secured, turning off outside lights and reviewing what happened while the user was away.
The interface covers devices available through the Google Home ecosystem. Current independent launch reporting cites Nest cameras and doorbells, Nest thermostats, Matter devices and other Works with Google Home equipment as examples.
Setup requires a Google Cloud project and OAuth
Home MCP is an authenticated remote service. Google’s documented setup flow requires users to create a Google Cloud project, enable the Home API, configure an OAuth consent screen and create a Web application OAuth client ID.
The OAuth redirect URI depends on the agent client. Google currently documents a callback for Antigravity, a Claude callback and a locally configured OpenClaw redirect URI. Users then authorize the selected Google Home structure through the Google permissions flow.
The production MCP endpoint documented for manual client configuration is:
https://home.googleapis.com/mcp
Google lists the Home platform OAuth scope as:
https://www.googleapis.com/auth/home.platform.v2
Client secrets should be handled as credentials. Operators using local agent gateways should keep them out of shared configuration exports, public repositories and logs.
Antigravity, Claude Cowork and OpenClaw
Google provides client-specific instructions for three agent environments.
Antigravity supports MCP configuration through its settings and uses Google’s OAuth callback. After adding the server, the user authenticates and selects the Home structure the agent can access.
Claude Cowork can add Home MCP as a custom connector. Google’s instructions use the Home MCP server URL plus the OAuth client ID and secret, followed by the Google authorization flow.
OpenClaw uses the remote SSE transport with OAuth. Google’s example configuration points OpenClaw at the Home MCP URL, supplies the Home platform OAuth scope and then uses openclaw mcp login home_mcp to start authorization.
MCP compatibility is the protocol boundary, so the service is not inherently limited to those three clients. They are the environments for which Google currently publishes setup guidance.
Safety controls and household data
Connecting an agent gives it meaningful access to a physical environment. Google explicitly warns that a connected agent can control devices and access home data, and that agent behavior can sometimes be unexpected or undesired.
Home MCP applies rate limits and safety protections. Google identifies unlocking doors as an example of a sensitive action that agents are prohibited from performing. Access can be revoked from the Google Home app or the user’s Google account.
Historical access also matters for privacy. list_home_history can expose state changes and event records, while resource and state tools expose the layout and current condition of connected equipment. Google recommends informing other household members when a shared home is connected to an agent. A separate development home is an available option for testing.
Familiar-face data has an additional consent boundary. Google requires explicit separate consent, manager status for the Home structure, a compatible Nest camera or doorbell, and Familiar face detection enabled on the relevant device.
Current limitations
Google lists two operational issues in the early-access documentation: some experimental traits may behave inconsistently, and Home MCP can experience higher-than-expected latency.
The server currently cannot create or manage Google Home automations. Google lists automation support as a coming feature. Device actions available through run_home_actions therefore represent a different capability from authoring persistent Home automation rules.
The launch is also access-constrained. The initial availability reported September 16 is tied to Google Home Premium Advanced in the U.S. Google has not provided a broader rollout date for other subscription tiers or markets.
Where Home MCP fits
Home MCP turns Google Home into an agent-accessible tool surface instead of requiring every AI client to implement a proprietary smart-home integration. For users already running personal agents, the practical gain is the combination of live telemetry, history and controlled actions through one MCP endpoint.
The strongest use cases are workflows that need context across several devices: diagnosing a device that went offline, reconstructing activity from event history, building a temporary dashboard, checking the state of a home during another task, or sending a completion notification through a supported speaker.
The security model makes client selection consequential. Google controls the Home-side authorization and action restrictions; the connected agent also receives the data needed to perform the requested workflow. Users should therefore grant Home MCP access only to an agent environment whose credential handling, data retention and tool-execution controls fit the household’s requirements.
Bottom line
Google Home MCP is a substantial expansion of the smart-home control surface for AI agents. Early-access users can give an MCP-compatible client structured access to devices, live state, historical events and supported actions through Google’s authenticated Home service.
The current entry requirements are specific: Google Home Premium Advanced, a Google Cloud project, Home API access, OAuth configuration and a compatible agent. Rate limits, sensitive-action restrictions, revocable access and separate familiar-face consent provide important control boundaries while Google expands the early-access implementation.
For agent developers and advanced smart-home users, the durable value is interoperability: one standard MCP interface can expose Google Home context to multiple agent environments without building a separate device integration for each client.
Sources
- Google Home Developers — Home MCP Server documentation, last updated September 15, 2026.
- TechCrunch — “Your AI agents can now control your Google Home devices,” September 16, 2026.
- 9to5Google — Google Home MCP early-access coverage, September 16, 2026.