Google opens Google Home to third-party AI agents via new MCP server

Google opens its smart home to outside agents

Google has started rolling out a Model Context Protocol (MCP) server for Google Home, giving third-party AI agents a standardized way to control connected devices and work with the data those devices generate. Announced on Tuesday, the integration lets any agent that speaks MCP plug into a Google Home setup and act on the user's behalf. The Verge names Claude, Google's Antigravity, Hermes and OpenClaw as compatible agents, while TechCrunch also lists ChatGPT.

The news came in a blog post from Taylor Lehman, group product manager for Google Home & Nest, according to The Verge. Notably, the integration does not replace Gemini for Home, which remains Google's own interface through the Home app and Nest speakers following the retirement of Google Assistant. Instead, it adds a parallel layer of control in which outside agents interact with devices through their own apps and interfaces.

What agents can actually do

In practice, the feature lets people use natural language to review camera summaries, monitor activity, control devices and build custom smart home dashboards, TechCrunch reports. The Verge offers more concrete examples: cross-camera analysis, such as asking an agent what happened after a child got home from school, or querying device state history to work out how many laundry loads were done last week or how long the lights were left on. Agents can also communicate by voice, for instance playing an audio message over a Google Home speaker when a task finishes.

Setup, devices and availability

Setup requires creating a Google Cloud project and configuring it to use Home MCP, then handing the configuration details to your chosen agent, which prompts a sign-in and permission grant. A setup guide is available in the Google Home Developer Center, TechCrunch says. Compatibility spans the whole ecosystem, including Nest doorbells and thermostats as well as Works with Google Home and Matter devices like light bulbs.

Availability is narrow at launch. Access is rolling out over the coming weeks to Google Home Premium Advanced subscribers in the United States, the $20-per-month or $200-per-year tier that TechCrunch notes includes longer event-based video history, descriptive notifications and searchable video history. Google would not comment on whether the MCP server will reach other subscription tiers or markets, and is collecting feedback from early adopters through its Smart Home for Developers Community.

Guardrails and caveats

Letting an agent touch smart locks, thermostats, HVAC systems and appliances raises obvious safety and privacy questions. The Verge reports that Home MCP enforces rate limits and safety protections, and will not allow an agent to unlock doors. Even so, Lehman cautioned that connecting an agent can produce surprising or unwanted behavior, and recommends reviewing Google's developer policies and terms of service.

Google already supports MCP elsewhere in its business, including Google Cloud, its data platforms, developer tools and Workspace, TechCrunch notes; the Home version is aimed more directly at consumers experimenting with agents for everyday tasks. The concept also has precedent: The Verge points to a similar MCP integration in the open-source Home Assistant platform, which its reporter used with Claude to troubleshoot devices, build automations and design dashboards in plain language — tasks that defeat most conventional smart home interfaces.

A developer play as much as a consumer feature

The Verge argues the consumer angle is only part of the story. Google has spent the last few years positioning itself as the smart home's underlying plumbing, opening API access in 2024 and expanding Gemini for Home into a complete AI offering. The MCP server extends that strategy: Google supplies the AI and smart home infrastructure while other companies build consumer-facing products on top, much as AWS became the foundation for a generation of internet and software businesses. The announcement also clears a path for developers to use Google's Antigravity coding tool to build custom smart home agents.

Whether developers commit is an open question. The Verge recalls Google's long record of launching and then abandoning smart home platforms, including Android @ Home, Weave, Project Brillo, Works with Nest and Google Assistant, and suggests developers will have to judge whether Google has finally settled on a platform it intends to keep.

Why it matters

MCP is rapidly becoming the standard bridge between AI agents and real-world systems, and the home is the most personal system most people own. By exposing event history and cross-device reasoning rather than simple on/off commands, this moves the smart home from command-and-control toward something genuinely proactive — a system that can troubleshoot, spot patterns and act. It also puts a subscription price on letting a third-party model see into, and act within, your house, sharpening questions about data, trust and physical safety. For developers, it signals that Google wants to be the platform layer for agentic smart homes, provided it sticks around this time.