Google Home opens the door to outside AI agents

Smart speakers have spent the last decade acting like glorified light switches. Google is finally trying to turn them into something that actually thinks.

On Wednesday, Google rolled out early access to a Model Context Protocol (MCP) server for the Google Home ecosystem. The move allows external AI agents—including Claude, ChatGPT, OpenClaw, Hermes, and Google Antigravity—to securely interact with connected smart home hardware and browse device event history.

From voice toggles to real agency

Instead of memorizing exact voice commands, users can now instruct their AI agents using plain conversational language. You can ask an agent to review security camera footage, summarize daily smart home alerts, trigger devices, or even construct a customized smart home dashboard from scratch.

Hardware coverage is surprisingly broad from day one. The integration works across the entire Google Home ecosystem, spanning first-party Google Nest doorbells and thermostats as well as third-party Matter and "Works with Google Home" devices like smart plugs and bulbs.

Setting it up, however, is not a simple one-click toggle. Users must set up a Google Cloud project, configure it to run Home MCP, hand the configuration details over to their chosen AI agent, and grant account permissions via an authentication flow documented in the Google Home Developer Center.

While Google has already rolled out MCP across Google Cloud, developer utilities, and Google Workspace, this marks the company’s first direct attempt to bring the protocol straight to consumer smart home hardware.

The $20 paywall

There is a catch: access is strictly gated for now.

The feature is rolling out exclusively to U.S. subscribers on the Google Home Premium Advanced tier, which runs $20 per month. That plan already bundles high-end perks like longer event-based video history, granular notification alerts, and natural language video search tools.

Google declined to comment on whether or when the integration will expand to cheaper subscription plans or international markets, noting only that it is collecting early impressions via its Smart Home for Developers Community.

Why it matters

This is a massive signal that smart home control is moving away from proprietary voice assistants and toward portable, model-agnostic AI agents. By choosing Anthropic’s open MCP standard instead of forcing everyone through Gemini, Google is acknowledging that power users already have their own preferred AI workflows and just want them connected to the real world.

We think this puts serious pressure on Apple and Amazon. HomeKit and Alexa have spent years locking users into rigid automation rules. Letting a third-party agent like Claude parse your security feed and make context-aware decisions completely changes the baseline expectation for home automation.

That said, the barriers to entry remain sky-high. Gating this behind a $20 monthly subscription and requiring a Google Cloud project keeps it firmly in developer territory for now. There is also the glaring question of physical security: handing agentic systems access to your front door and cameras carries real risk if an agent hallucinates or falls victim to prompt injection. Google’s slow, gated rollout is likely less about monetizing early and more about ensuring things don't go haywire in the living room.