Google Launches Home MCP, Letting AI Agents Like Claude and Antigravity Control Smart Home Devices

Google today launched Home MCP, a Model Context Protocol server that gives AI agents direct access to smart home devices and event history across the Google Home ecosystem. The rollout, announced September 16, 2026, targets Google Home Premium Advanced subscribers in US English.

Home MCP works with any agent that supports calling MCP tools, including Google Antigravity, Anthropic's Claude, Hermes, and OpenClaw, according to Google. Once connected, an agent can query and act on Nest cameras, Nest doorbells, Nest Thermostats, Matter-compatible light bulbs, and other devices registered in a user's Google Home setup.

What agents can do

Google outlined several use cases for the integration:

• Cross-camera analysis: Ask an agent what happened when the kids got home from school, and it pulls a summary with clips from multiple cameras.
• Historical queries: Agents can parse device state history to answer questions like how many laundry loads ran in a week or how long lights stayed on.
• Voice responses: An agent completing a background task can send an audio update through a Google Home speaker.
• Custom dashboards: Agents can generate a tailored smart home dashboard using natural language requests.

Google describes the goal as giving agents "real-world physical context" so they can reason about home activity without a developer building custom integrations for each device type.

Security guardrails

Google states that Home MCP "enforces rate limits and safety protections, such as prohibiting sensitive actions like unlocking doors." The company also warns that behavior can vary by agent: "depending on your agent, connecting it to Home MCP can result in unexpected or even undesired behavior," and it advises users to review third-party developer policies and terms of service before granting access.

Setup process

Home MCP requires users to create a Google Cloud project and configure it for the service. From there, users hand the MCP configuration details to their agent of choice and ask it to complete setup, which triggers a sign-in and permissions flow. Google has published full setup guides on its Home Developer Center.

Pricing for Home MCP access itself was not disclosed beyond the requirement of a Google Home Premium Advanced subscription; Google did not specify separate MCP usage fees in its announcement.

What this means

Home MCP is Google's clearest move yet toward treating the smart home as an addressable surface for third-party AI agents rather than a walled garden tied to Google Assistant or Gemini alone. By adopting the open Model Context Protocol — the same standard Anthropic introduced and that has since been picked up across the industry — Google is betting that interoperability will drive more usage of Google Home hardware, even if the agent doing the work is Claude or a competitor's product rather than Google's own.

The restriction against sensitive actions like unlocking doors signals Google is aware of the liability exposure in letting general-purpose agents control physical infrastructure, particularly given the explicit warning that agent behavior can be unpredictable. The limited initial rollout — Premium Advanced subscribers, US English only — suggests Google is treating this as a controlled pilot before wider availability. Expect scrutiny over how camera footage and event history get exposed to third-party agent providers, since Home MCP hands a potentially large volume of personal video and behavioral data to external AI systems by design.