Product1 publisher2 min readPublished
Google now lets Premium Advanced subscribers point Claude at their Home device graph
The new Model Context Protocol endpoint reaches every device and the whole event history in Google Home, but only for US subscribers on the paid Premium Advanced tier, and Google is shipping it as early access.
The Product Desk · Product desk
What happened
- Google told Home customers and developers on Wednesday that AI agents other than its own Gemini will be able to control smart home devices, using Model Context Protocol tool calls.
- The announcement names Anthropic's Claude, the open-source agents OpenClaw and Hermes, and Google's Antigravity coding platform as agents that will soon work with Google Home.
- Access began rolling out on Wednesday to Google Home Premium Advanced subscribers in the US only, on a $20 subscription, as early access Google says may be buggy or limited.
- Google says the agents will handle more complex tasks, including compiling text summaries of what security cameras have seen.
- CNET reports that Google Home is the first of the big three home platforms, alongside Apple Home and Amazon Alexa, to offer anything like this.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- cost A household that wants Claude instead of Gemini pays Google $240 a year for the connection, on top of whatever it pays for the agent itself.
- exposure The grant Google describes covers the whole ecosystem, so a household approving one agent puts its camera and sensor timeline within reach of that agent, not just a thermostat.
- constraint US-only early access means an installer or integrator cannot offer this as a supported flow to customers elsewhere, or promise it will behave the same next month.
- precedent Home Assistant has run these agent setups for years without a subscription. Buyers get a price comparison the moment Apple or Amazon ships its own version.
A household needs two things: a US billing address on Google Home Premium Advanced, and an agent that already calls Model Context Protocol tools [4][1]. Google started the rollout on Wednesday and called it early access, so bugs and limitations are expected [4].
Google's announcement, as quoted by CNET, said that any agent supporting Model Context Protocol tool calls can now interact with all of the devices and event history in the Google Home ecosystem [2]. Devices and history, ecosystem-wide. Google had already used MCP this way to let people add agents to Workspace [6].
Twelve months of Premium Advanced comes to $240 [15]. What that buys is a route into devices the household already owns and an event log it already generated [5]. Google did not describe this as retiring Gemini for Home, which it still ships as the voice assistant and controller for the same devices [1][5].
CNET's reviewer reported two separate faults with Gemini for Home: it does not always understand complex commands as intended, and it sometimes fails to operate devices that are connected and ready [8]. Those want different fixes. A different model can parse the sentence. The second fault sits below the agent, in the device integration itself, which every MCP client will call the same way, so a household that switches to Claude and still has a light that will not come on has changed the caller.
SwitchBot is building proprietary hubs so its own hardware talks natively to OpenClaw, and Homey sells universal hubs for mixed devices [12][13]. Both are doing per-agent hardware work, still under way. On the Google side, CNET describes the third-party agent as an extra layer of control on top of Google Home's existing device support and automations [16].
For whoever has to answer for this on Friday, the question is whether last month's failures happened before the device was called or after. A new agent may help with the ones that happened before; for the ones that happened after, MCP changes nothing. The second question is who owns the custom dashboard Google is offering in place of its own interface [7]. CNET's reviewer, who has used builds like it, wrote that they are "still a bit clunky, get things wrong, don't always do what they say and are difficult to troubleshoot thanks to the vibe coding involved" [9]. Google is directing early-access feedback to its smart home developers' help channel [14]. A Google representative did not immediately respond to CNET's request for comment [17].
What to watch
- Whether Google extends the MCP endpoint below the Premium Advanced tier or outside the US.
- Whether Apple Home or Amazon Alexa publishes an equivalent agent endpoint.
- Whether Google adds scoping controls so an agent can be granted some devices or some event history rather than the whole ecosystem.