Google Home as open infrastructure — MCP lets Claude, ChatGPT, and any AI assistant control your Nest

Google has built a public Model Context Protocol server for Google Home, and the implications extend beyond convenience: the decision to open Nest hardware to competitor AI assistants signals a deliberate choice to treat smart home control as shared infrastructure rather than a Gemini advantage.

The devices that previously only answered to Gemini now answer to Claude, ChatGPT, Hermes, OpenClaw, and any other AI built on the same standard. Google has built a public Model Context Protocol server for Google Home, and as of September 16, 2026, any AI assistant that supports MCP can authenticate against it and issue commands to whatever Nest hardware is registered to an account. Thermostats, cameras, lights, locks, smart plugs, speakers: the full scope of what a user would access through the Google Home app is now reachable through a conversational AI, whichever one they prefer.

Model Context Protocol — MCP — is the specification that defines how AI agents connect to external tools and services. Anthropic introduced it as an open standard in November 2024, and in the roughly two years since, it has become the closest thing the AI industry has to a shared communication layer for AI-to-tool integration. The model is analogous to how HTTP standardized web communication: instead of every AI application building one-off integrations with every service, MCP provides a common interface that both sides can implement once and reuse everywhere. Google is not the first company to build an MCP server — the ecosystem already includes integrations across productivity tools, databases, and development environments — but in terms of consumer device reach, the Google Home launch is by far the most significant entry yet.

The capabilities available through the Google Home MCP connection mirror what users already access through the Google Home app. On the control side: adjusting thermostat schedules, toggling smart plugs, controlling lights by room name or individual device. On the read side: pulling camera event history, monitoring smart home activity, checking device states. Support extends across Nest thermostats, Nest cameras and doorbells, connected speakers and displays, and Matter-enabled devices — the standard that covers an increasingly broad slice of third-party smart home hardware. The interface is described as complete and functional, not a demonstration subset intended to show off the integration while limiting real utility.

Access launched on September 16 in early access form, currently restricted to Google Home Premium Advanced subscribers in the United States. That subscription costs $20 per month — the same tier that already includes expanded camera history, unlimited event storage, and Nest Aware Plus features. Setup requires creating a Google Cloud project and completing an MCP permissions grant through the Google Home app; it is not a one-click enable, but it is not a developer-facing API requiring code either. The restriction to the Advanced tier reads as a gating mechanism for the beta rather than a permanent pricing decision. Google has not published a timeline for broader availability, but the structure of the launch — an explicit early access phase with no announced endpoint — is consistent with a controlled rollout ahead of general availability.

Four AI assistants were available to connect on launch day: Claude (from Anthropic), ChatGPT (from OpenAI), Hermes, and OpenClaw. Users initiate the connection from within the Google Home app, selecting which assistant to authorize and which categories of devices to grant it access. The authorization granularity matters here: a user can authorize Claude to control lights and thermostats while restricting it from accessing camera history, for instance, and separate authorizations for different assistants do not cascade. Connecting Claude grants Claude the permissions explicitly assigned to Claude; it does not extend those permissions to ChatGPT or any other assistant on the account.

That per-assistant, per-scope authorization model is not incidental — it reflects the way MCP was designed to handle trust in multi-agent environments. Each AI assistant operates within the permissions explicitly granted to it for that connection, and the Google Home app serves as the authorization surface where users manage those grants. In practice this means a household that uses multiple AI assistants can give each a different level of access depending on how much they trust it with specific hardware categories, and revoke those grants individually without affecting other connections.

The decision to open Google Home to competitor AI assistants is the most strategically interesting part of this announcement. Google makes Gemini. Gemini has controlled Google Home natively for years. Choosing to build a public MCP server that Claude and ChatGPT can use is a deliberate decision to stop treating Gemini’s home automation capabilities as a proprietary advantage and start treating Google Home hardware as platform infrastructure. The reasoning is probably about ecosystem momentum: smart home hardware is a durable, high-investment category. The households that treat a smart home as a serious investment — not a novelty — are the ones who buy more hardware, subscribe to services that manage it, and stay within a compatible ecosystem for years. A Google Home that works fluently with whatever AI assistant a household prefers is a more compelling hardware purchase than one that only works fully with Gemini. Google is betting that ecosystem openness generates more hardware revenue and subscription stickiness than exclusivity would.

For the AI assistant providers, the calculus is more straightforward. Access to over 100 million installed Google Home households on day one is a substantial expansion of what each assistant can actually do. An AI assistant that can only answer questions occupies a narrower utility space than one that can execute tasks in the physical environment where a user lives. The practical scope of what it means to have an AI assistant has shifted: asking Claude to turn off the lights and drop the thermostat before bed is now a functional capability rather than a hypothetical one, for any Claude user with a Nest device and a Premium Advanced subscription.

MCP’s broader trajectory is worth holding alongside this specific launch. The protocol that started as an Anthropic-published open standard in late 2024 has moved from developer tooling into consumer infrastructure in under two years. The Google Home integration is one of the most visible examples of MCP crossing that line — from a technical specification most users would never encounter into a feature that changes what a smart speaker or thermostat can do in practice. How quickly the remaining AI platforms and smart home ecosystems follow this pattern is an open question, but the pressure from Google’s participation is significant. If the largest consumer smart home platform treats MCP as the default integration surface, other platforms face a compatibility gap that grows harder to justify leaving open.

The $20 threshold and the US-only gate will limit who actually uses this in its current form. But the launch period is rarely the meaningful measure of a platform change — what matters is whether the architectural decision holds and broadens. Google has chosen open infrastructure over lock-in. That choice, if it sticks, reshapes what an AI assistant actually is in a domestic context.

Tags: AI agents, Claude, chatgpt, Google Home, MCP, Nest