Google will let any AI agent run your home
Google is opening its smart home ecosystem to third-party AI agents, allowing tools such as Claude and Open Claw to analyse data and control connected devices through a new integration.
Google is giving artificial intelligence agents a new role in the smart home, allowing third-party AI tools to access and control connected devices through a new system designed to make homes more automated.
The company has introduced Google Home Model Context Protocol (MCP), an integration that allows AI agents supporting the standard to work with devices and activity data connected to a Google Home ecosystem.
The technology means tools including Anthropic’s Claude, Google Antigravity, Hermes and Open Claw can interact with smart home devices, analyse events and carry out tasks on behalf of users.
Taylor Lehman, group product manager at Google Home and Nest, said the integration could allow AI agents to understand what is happening inside a home and provide more useful assistance.
Potential uses include analysing security camera footage to answer questions about events at home, reviewing device history to estimate energy use or household activity, and creating personalised dashboards for controlling smart devices.
AI agents could also communicate through Google Home speakers, such as sending an audio message when a requested task has been completed.
The move does not replace Google’s Gemini for Home assistant, which remains the company’s main interface for interacting with Google Home devices through the Home app and Nest speakers.
Instead, MCP gives other AI systems a way to directly connect with smart home devices.
At launch, access is limited to Google Home Premium Advanced subscribers in the US, with availability expanding in the coming weeks.
Users will need to create a Google Cloud project and configure it to work with Home MCP.
However, giving AI agents access to real-world devices has raised concerns about privacy and security.
Smart locks, heating systems and household appliances could become vulnerable if an AI system behaves unexpectedly or is compromised.
Google said Home MCP includes safety protections and limits on how often actions can be performed, and the company said agents will not be able to unlock doors through the system.
Lehman also warned that connecting external AI agents to Home MCP could result in “unexpected or even undesired behavior”, advising users and developers to review Google’s policies before using the technology.
