Google is opening its smart-home platform to a wider range of AI assistants, allowing third-party AI agents like Claude and Antigravity to access and control connected devices through a new Model Context Protocol (MCP) integration.

The move has the potential to significantly broaden how consumers interact with smart home products. The experience could shift the experience from simple individual voice commands towards AI agents carrying out more complex tasks across multiple devices in the home.

Google’s Home MCP server is currently available in early access and provides a bridge between Google Home and compatible AI agents. Through the system, AI assistants can discover connected devices, check their status, access historical activity and issue commands.

The technology means an AI assistant could potentially respond to requests like checking if lights have been left on, reviewing home activity via security cameras while a homeowner is away or controlling other compatible connected devices.

Google is not limiting the capability to its own Gemini platform. According to the company, it will be compatible with a range of MCP-enabled AI clients, including Anthropic’s Claude and Google Antigravity. Other compatible agents can also connect to the platform.

At the same time, Google is expanding Gemini’s role within Google Home. The company has made improvements to language controls and AI-powered features across its smart home platform.

The MCP integration, however, takes things a step further by allowing third-party AI agents to actually interact with Google Home devices and data.

The development highlights a move towards a more agent-driven smart home model, in which AI is an intermediary between users and their connected devices. Rather than users operating a number of separate applications or making individual commands, AI agents could potentially coordinate across multiple devices and interpret information from around the home.

Google has also introduced restrictions around what agents can do. The company says certain sensitive actions, including unlocking doors, are blocked and there are limits on how frequently AI agents can interact with connected home devices. Google also warns users that AI agents can produce unexpected results when given access to real-world devices.

Home MCP is currently an early-access service and requires a Google Home Premium Advanced subscription, a Google Cloud project and a compatible MCP client.

Google says additional capabilities are also in development, including support for creating and managing smart-home automations through MCP.

It’s a step towards AI becoming a central interface for connected-home products and moves Google Home from a platform for Google’s own technology to infrastructure that can be accessed by a broader AI ecosystem.