Google Home opens its doors to third-party AI agents through MCP
Google Home now lets third-party AI agents control connected hardware and analyze household data through the Model Context Protocol, giving tools from Anthropic and others a direct line into the smart home’s device layer. The integration, announced on September 16, 2026, is positioned as an open entry point for any MCP-compatible agent rather than a replacement for Google’s own Gemini for Home assistant.
What the new integration actually does
Home MCP, as Google calls it, exposes a smart home’s devices and event history to external agents in a standardized format. According to Taylor Lehman, group product manager at Google Home and Nest, the integration “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.”
That access unlocks several capabilities that go beyond simple voice commands:
- Cross-camera analysis. An agent can pull footage from multiple cameras to answer a question such as what a child did after arriving home from school.
- Device state history. An agent can review past activity to count loads of laundry done in a week or measure how long lights were left on.
- Voice messaging. An agent can send an audio message through a Google Home speaker when it has finished a task.
- Custom dashboards. An agent can generate a tailored control interface for managing the home.
How it fits with Gemini for Home
Home MCP does not replace Gemini for Home, which remains the primary interface for Google Home users following the retirement of Google Assistant. Google Nest speakers continue to respond to Gemini for Home commands. What MCP adds is a parallel lane: third-party agents can reach Google Home devices and data through their own applications and user interfaces, without going through Google’s assistant.
This split matters because it signals where Google sees its competitive advantage sitting. Voice assistance is one product; the infrastructure that lets any agent read home data and trigger devices is a separate, more durable layer.
Who can use it, and how to get started
At launch, access is limited to Google Home Premium Advanced subscribers in the United States, priced at $20 per month or $200 per year. Availability is rolling out over the coming weeks.
Setup is more involved than installing a typical smart home skill. Users need to create a Google Cloud project and configure it to use the Home MCP. That step alone puts the feature outside the reach of casual users and frames it, at least initially, as a tool for developers and serious tinkerers.
Safety and privacy limits baked into the system
Granting an AI agent access to physical hardware raises immediate questions about what it can and cannot do. Google has built several guardrails into Home MCP:
- Rate limits apply to agent requests.
- Certain actions are blocked outright, such as unlocking doors.
- Lehmant’s blog post warns that, depending on the agent chosen, “connecting it to Home MCP can result in unexpected or even undesired behavior,” and points users toward Google’s developer policies and terms of service.
The restrictions do not eliminate risk. An agent that can read camera feeds, adjust a thermostat, or trigger routines still operates in a space where mistakes have physical consequences. Reviewing which agent gets access, and what permissions it carries, becomes part of running the system.
Why this matters beyond turning lights on and off
The headline feature is convenient, but the deeper change is structural. A traditional smart home responds to commands: a user taps an app or speaks a phrase, and a single device reacts. An agent connected through MCP can read the home’s history, reason across multiple devices, and take actions on the user’s behalf. That moves the smart home from command-and-control toward something contextual and proactive.
Home Assistant, the open-source smart home platform, implemented a similar MCP integration earlier, and early testing showed what is possible. An agent can troubleshoot a misbehaving device, suggest improvements to an automation, design a custom dashboard, and set up configurations that would normally require navigating multiple menus, all through natural-language requests.
For consumers, this is the practical payoff: a system that can identify patterns across the home is more useful than one that responds with “I’m sorry, I can’t do that.”
Google’s bigger play: becoming the smart home’s infrastructure
Google has spent several years repositioning itself as the underlying platform for the smart home rather than just one of its consumer brands. The company launched API access to its smart home in 2024 and more recently turned Gemini for Home into a full-stack AI offering. Home MCP extends that strategy by inviting external agents to build on top of Google’s device layer.
The business model is reminiscent of how Amazon Web Services became the infrastructure beneath a generation of internet companies. Google does not need every consumer to use a Google-branded assistant; it needs developers and AI companies to choose Google Home as the place their agents go to act on the physical world. Whether that vision holds depends on whether developers trust Google to keep the platform stable. The company has launched and retired several smart home programs over the years, including Android at Home, Weave, Project Brillo, Works with Nest, and Google Assistant. Each shutdown made the next launch harder to sell.
Google has also opened a path for developers to use its agentic Antigravity coding tool to build custom AI agents that tap into Google Home device data and control layers. Even if those developers use a non-Google agent at the user-facing end, the underlying home data and devices still flow through Google’s infrastructure.
FAQ
What is Google Home MCP?
Home MCP is a new Google Home feature built on the Model Context Protocol. It lets third-party AI agents that support MCP, including Claude and Open Claw, read your Google Home device data, review event history, and control connected devices through their own applications.
Who can use Google Home MCP right now?
At launch, the feature is available only to Google Home Premium Advanced subscribers in the United States, with the $20-per-month or $200-per-year plan. Access is rolling out over the coming weeks, and setup requires creating and configuring a Google Cloud project.
What can and cannot a third-party agent do through Home MCP?
Agents can analyze data across cameras, review device state history, send voice messages through Google Home speakers, build custom dashboards, and control many connected devices. Google enforces rate limits and blocks certain actions, such as unlocking doors. The specific behavior of any agent also depends on that agent’s own policies, which Google recommends reviewing before connecting.
Related coverage
This article summarizes reporting from theverge.com.