Google Home Opens Up to AI Agents: Early Access to Home MCP
Google has opened early access to Home MCP, allowing any AI agent that speaks the Model Context Protocol (MCP) to read and control a Google Home. Agents like Claude, OpenClaw, Hermes, and Antigravity are among the first to gain access.
Key Features:
- Summary footage from cameras across your home.
- Access to device state history, such as laundry loads or light usage.
- Audio messages pushed through Google Home speakers after task completion.
- Custom dashboards generated by the agent, replacing the standard Google Home app interface.
Setup and Requirements:
- Create a Google Cloud project and enable the Home API.
- Configure OAuth consent screens and obtain client ID and secret.
- Specify the redirect URI for the agent and publish the application.
- Currently available only for Google Home Premium Advanced subscribers in the United States (English).
Restrictions:
- No automation support yet.
- Known issues include latency and experimental device traits.
- Sensitive actions are blocked, such as unlocking doors.
- Facial recognition requires separate consent and setup on compatible Nest cameras or doorbells.
- Users must inform other household members about the testing environment.
Security Concerns: Prompt injection is a known vulnerability where malicious instructions can be hidden within regular content, executed by the agent.