Google began an early access rollout of a Model Context Protocol server for Google Home on September 16, letting AI agents control connected devices and review event history through natural language. The company said so in a post covered by TechCrunch and Engadget.
Google named Claude, ChatGPT, Hermes, Open Claw and its own Google Antigravity as agents that can already connect, per TechCrunch. Supported devices include Nest doorbells and thermostats and any "Works with Google Home" or Matter-compatible device, such as smart bulbs.
Access at launch is limited to U.S. subscribers of Google Home Premium Advanced, a $20-a-month tier, and requires setting up a Google Cloud project and granting permissions before connecting an agent, TechCrunch reported. Google has not said when the feature reaches other subscription tiers or countries.
Anthropic introduced MCP in 2024, and Google, OpenAI, Microsoft and SpaceXAI now support the standard, Engadget reported, though the outlet noted security researchers have flagged authentication gaps and data breaches tied to MCP adoption more broadly.
Letting a chat agent open a smart lock or view a camera feed is exactly the kind of task MCP was built for, and exactly the kind where a compromised agent or a leaked token stops being a data breach and starts being a break-in.