Google opens its smart home to AI agents behind a $20 subscription

Google has opened Google Home to outside AI agents. The company rolled out early access to a Model Context Protocol server that lets an agent inspect your home structure, read live device states, run commands and query event history.

Access is narrow at launch. TechCrunch reported that it’s limited to Google Home Premium Advanced subscribers in the US, arriving over the coming weeks, and that Google wouldn’t comment on whether other subscription tiers or markets follow. Engadget put the price of that tier at $20 per month or $200 per year.

Allows any AI agents that support MCP, including Google Antigravity, Claude, Hermes or Open Claw, to securely work with all of the devices and event history in your Google Home ecosystem.

Taylor Lehman, group product manager, Google Home & Nest, via The Verge

That matters because of what sits underneath the phrasing. MCP is an open standard for connecting AI applications to external systems, and Google’s developer documentation spells out exactly which five tools the Home server hands an agent.

Tool	What it does
list_homes	Retrieve accessible homes and structures
list_home_resources	List devices, area layouts, traits and command schemas
list_home_states	Check real-time connectivity and trait states
run_home_actions	Execute parameterised action commands on devices
list_home_history	Query past state changes and event logs over a time range

The catch is that none of this is a toggle in an app. Google’s setup guide asks you to create a Google Cloud project, enable the Home API, configure an OAuth consent screen, generate a client ID and secret, then publish the app. Redirect URIs are documented for Google Antigravity, Claude Cowork and OpenClaw.

Google is blunt about the risk in the same page. The documentation warns that Home MCP enforces rate limits and safety protections, including a block on sensitive actions such as unlocking doors. Even so, it adds, the outcome depends on which agent you point at it.

Depending on your agent, connecting it to Home MCP can result in unexpected or even undesired behavior.

Google Home Developers, Home MCP documentation

Some data stays behind a second gate. Familiar faces from Nest cameras and doorbells need separate explicit consent from a structure manager, granted through a per client consent link. An agent’s access can be revoked from the Google Home app or your account page at any time.

The sceptical reading writes itself. Engadget notes that experts have flagged serious security flaws in MCP, including a lack of enforced authentication, which is the sort of gap we covered when arguing that agents are dangerous enough to matter before they’re reliable enough to trust. A thermostat is a lower stake than a database, but a camera history isn’t.

There’s a second objection, and The Verge makes it plainly: Google has launched and buried smart home platforms before, naming Android @ Home, Weave, Project Brillo, Works with Nest and Google Assistant. Developers deciding whether to build on Home MCP are being asked to trust a track record, not a roadmap. The permission plumbing here looks a lot like the governance layer enterprises are buying for their own agents.

What’s worth watching is the gap Google has already admitted. Creating and managing automations isn’t supported yet, some device traits are marked experimental, and the docs list latency as a known issue under active optimisation. Until automations land, an agent can read your home and issue commands, but it can’t build the rules that run it while you’re out.