Google Home Welcomes Third-Party AIs via MCP Protocol

Key Takeaways

1Google Home opens access to third-party AI agents via the MCP protocol

2The rollout begins in early access for Premium Advanced subscribers in the United States ($20)

3Agents like Claude, OpenClaw, Hermes, and Antigravity will be able to control devices, generate summaries, and use custom dashboards

4Security risks, such as prompt injection, have been identified

Why it matters : Google Home becomes the first of the major smart home platforms to allow the integration of third-party AI, paving the way for advanced uses while raising new security concerns.

Google allows AI agents other than Gemini to access devices and Google Home history via the MCP protocol. The rollout begins in early access for the Premium Advanced offering in the United States, with a subscription of $20. Use cases such as camera summaries and custom dashboards are announced, while risks like prompt injection remain to be considered.

Early Access, Subscription, and Security Precautions

Support for third-party AI agents is starting to roll out for Google Home Premium Advanced subscribers in the United States. It is offered in early access, which may include bugs or limitations, and is reserved for users with a $20 subscription. Google encourages users to provide feedback through its smart home developer support channel. Support via MCP includes protections that Google can enforce within its ecosystem, but the opening to additional agents introduces new variables and the potential for vulnerabilities. Prompt injection, also known as smart home promptware, involves hiding malicious instructions in emails or calendar events to potentially take control of devices without the user's knowledge. Google has encountered similar issues in early versions of Gemini, which were quickly addressed. However, the addition of other AI systems could multiply the opportunities for such vulnerabilities to arise. A Google representative did not immediately respond to a request for comment.

What Compatible Agents Will Be Able to Do in Google Home

Agents supporting the tools of the Model Context Protocol will be able to interact with all devices and the event history of Google Home. Google announces capabilities such as generating text summaries from security camera feeds and using custom dashboards to control the home instead of the Google Home interface. For example, it will be possible to use Claude's voice mode to adjust a thermostat. These third-party agents are also presented as being able to set up additional automations and access data not provided by Gemini for Home alone.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

A First in Major Ecosystems, but Not in Home Automation

Solutions like Homey Pro have supported third-party AIs for years, as has the open-source platform Home Assistant, which is used to integrate advanced agents like OpenClaw and Claude, backed by a large library of guides and aimed at privacy-conscious users who prioritize local control. On the hardware side, universal hubs like those from Homey facilitate integration through plug-and-play compatibility, while manufacturers like SwitchBot develop proprietary hubs to ensure native communication with OpenClaw. Among major consumer ecosystems, Google Home becomes the first of the three platforms—Apple Home, Amazon Alexa, and Google Home—to offer similar support for third-party AIs. This is also the first time that Google Home users can add an additional layer of control with a third-party AI while maintaining support for existing devices and automations already in place.

The Technical Framework and Actors Announced by Google

Google has announced the addition of support for AI agents other than Gemini AI. Notable mentions include Claude from Anthropic, the open-source agents OpenClaw and Hermes, as well as Google's Antigravity coding platform. This openness relies on the Model Context Protocol, already used to add agents to Google Workspace, and whose deployment in Google Home has begun. It extends Gemini for Home, the voice assistant and controller already in place to manage lights, heating, security cameras, and thermostats compatible with the Google Home app.

Advanced Automations Considered and Areas for Improvement

Support for third-party agents is presented as offering potential improvements in understanding complex commands and managing connected devices. Advanced automation scenarios are envisioned, such as linking a thermostat's behavior to very specific local weather reports using versatile agents like OpenClaw. In contrast, creating custom dashboards via AIs like Claude is described as cumbersome, error-prone, and difficult to troubleshoot due to the coding involved.