Google opens Google Home to third-party AI agents with Home MCP

17 Sep 2026 4 min read 2845 views Technology
Technology

Google began early access to Home MCP on September 16, 2026, an official Model Context Protocol server that lets third-party AI agents — such as Claude, Antigravity, Claude Cowork, or OpenClaw — discover home devices, read their state and history, and execute authorized commands across the Google Home ecosystem. Until now, that capability was reserved for Gemini for Home; with Home MCP, users can choose which AI assistant connects to their home.

Initial access is limited to the United States and requires a Google Home Premium Advanced subscription ($20 a month or $200 a year). According to Google's developer documentation, a compatible agent can enumerate accessible homes, list rooms and devices along with their capabilities and available commands, query past event history, and execute authorized actions — for example, checking whether a light is still on, reviewing recent front-door activity, or asking how many laundry loads finished in a week.

Weekly context

According to XenoSpectrum, ResTech Roundup, KL Gadget Guy, and Newsy Today, Home MCP exposes its capabilities in several well-defined categories: list_homes to enumerate accessible homes and list_home_resources to return devices, rooms, traits, and available commands, plus real-time state reads and past-event queries. Access to facial recognition data for registered individuals — available only with compatible Nest cameras or doorbells — requires separate, explicit consent that only the home's administrator can grant. Actions on sensitive devices, such as locks, are directly excluded from agents' scope.

What changed

  • The connected home is no longer a Gemini-only walled garden: for the first time, AI agents outside Google can inspect and control Google Home devices with explicit user authorization.
  • Layered access, not all-or-nothing: Home MCP distinguishes between state reads, history queries, and command execution, with specific exclusions for high-risk actions like unlocking doors.
  • Reinforced consent for biometric data: facial recognition of specific individuals requires separate, explicit authorization from the home's administrator, plus compatible Nest hardware.
  • Paid subscription and limited availability: early access requires Google Home Premium Advanced and, for now, is only available in the United States.
  • Rate-limited calls and revocable access: Google applies rate limits to agent queries and lets users cut the connection at any time.

Impact for development and product teams

For developers building AI agents or assistants, Home MCP opens an official, standardized path to integrate with one of the world's largest connected-home ecosystems, without relying on Google-specific proprietary APIs. For product teams in the home automation and IoT space, the arrival of an MCP standard for the home reinforces the trend — already visible in Home Assistant and other platforms — toward open connection protocols between AI agents and physical devices, rather than closed per-manufacturer integrations. The layered permission design (reads, history, commands, biometric data) is also a reference pattern worth considering when designing the permission model for any AI agent integration with physical devices.

Practical recommendations

  1. If you build connected-home integrations, review the official Home MCP documentation and assess whether list_homes and list_home_resources cover your use cases before building a proprietary integration.
  2. Design any AI agent integration with physical devices by explicitly separating state reads, history queries, and command execution, following Home MCP's layered permission model.
  3. If your product handles biometric or facial recognition data, always require explicit consent separate from other permissions, as Google does in this case.
  4. Factor in the cost of the Google Home Premium Advanced subscription if your product depends on this integration, since it isn't free.
  5. Watch Home MCP's geographic expansion beyond the United States before planning an international launch that depends on it.

What to watch next

  • Whether Google expands Home MCP to more countries beyond the United States in the coming months.
  • New compatible AI agents joining the initial list (Claude, Antigravity, Claude Cowork, OpenClaw).
  • Whether Google eventually relaxes the exclusion of sensitive actions like lock control.
  • How other connected-home platform makers (Amazon, Samsung SmartThings, Home Assistant) respond to this move by Google.
  • The first real-world use cases documented by users and developers after the early-access period.

Conclusion: With Home MCP, Google turns the connected home into a space open to third-party AI agents, not just Gemini, applying a layered permission model that carefully separates state reads, history, and command execution, and that requires additional explicit consent for biometric data. It's a move worth watching both for agent developers and for any team designing AI integrations with physical devices.

Sources and documentation

Share