On September 16 2026, Google announced that its smart-home platform will accept external AI agents through a new Model Context Protocol integration, letting compatible systems such as Claude or Open Claw read device data, issue commands, and speak via Nest speakers while coexisting with the existing Gemini for Home assistant.
Short answer: On September 16 2026, Google announced that its smart-home platform will accept external AI agents through a new Model Context Protocol integration, letting compatible systems such as Claude or Open Claw read device data, issue commands, and speak via Nest speakers while coexisting with the existing Gemini for Home assistant.
Google announced on September 16, 2026 that its smart home platform will now accept external AI agents through a new Model Context Protocol integration. The feature, called Home MCP, lets any compatible AI system, such as Claude, Open Claw, Google Antigravity or Hermes, connect to a user’s Google Home ecosystem, read device data, issue commands and even speak through the home’s speakers. According to Google, the integration does not replace the existing Gemini for Home assistant; instead it adds a parallel layer that lets third-party tools work alongside the built-in voice assistant.
The MCP link gives agents access to the full history of events and the current state of every connected gadget. Users can ask an agent to review footage from multiple cameras to see what a child did after school, or to calculate how many laundry loads were run last week by examining washer cycles. Agents can also generate personalized dashboards that surface controls for lights, thermostats and locks in a single view, all driven by natural language requests. Voice feedback is possible too; an agent can send an audio message over a Nest speaker when it finishes a task, such as confirming that a door has been locked.
Initially, the MCP capability is limited to subscribers of Google Home Premium Advanced in the United States. The plan costs twenty dollars per month or two hundred dollars per year, with a broader rollout expected in the coming weeks. To enable the feature, users must create a Google Cloud project, link it to their Home account and activate the MCP setting within that project. Google says the process is aimed at developers and power users who want to experiment with custom AI-driven home automation.
Safety measures are built into the integration. Google enforces rate limits to prevent excessive commands and blocks certain actions, such as unlocking smart doors, to reduce risk. Nevertheless, the company warns that connecting an external agent can produce unexpected or undesirable behavior, depending on how the agent interprets data or executes instructions. Users are encouraged to review Google’s developer policies and terms of service before granting broad access.
The Verge notes that a similar MCP implementation already exists in the open-source Home Assistant platform, where the writer tested Claude to troubleshoot devices, redesign automations and build dashboards using plain language. That experiment showed that an agent with deep access to home data can handle complex tasks that typical apps struggle with, such as identifying patterns across multiple sensors or suggesting energy-saving tweaks based on historical usage.
For people who build AI applications, the announcement opens a new distribution channel. Developers can now target a large base of Google Home owners without having to create their own hardware or voice stack. By adhering to the Model Context Protocol, they gain a standardized way to read device states, issue commands and receive event streams, which simplifies cross-platform work. The requirement to set up a Google Cloud project may deter casual hobbyists, but it also provides a clear path to production-grade deployments, including billing, logging and access control.
For everyday users who rely on smart home conveniences, the change promises a shift from simple command-and-control toward more proactive, context-aware assistance. Instead of memorizing specific phrases for each device, users could ask a single AI agent to manage routines, diagnose faults or optimize energy consumption based on the home’s actual behavior. The ability to generate custom dashboards means that families can tailor controls to their own preferences, surfacing the most relevant information without digging through multiple apps.
However, the added convenience comes with heightened responsibility. Granting an external AI the ability to read sensor histories and actuate locks, thermostats or appliances raises privacy and security concerns. While Google’s safeguards block the most dangerous actions, the possibility of unintended behavior remains, especially if an agent misinterprets a command or acts on stale data. Users should therefore start with limited permissions, test agents in a controlled environment and monitor logs closely before expanding access.
In summary, Google’s Home MCP integration marks a notable step toward treating the smart home as a programmable environment that external AI can reason about and act upon. Developers gain a new avenue to deploy their models, while users may enjoy more intuitive and adaptive home management, provided they weigh the benefits against the need for careful oversight and robust security practices.
Google announced the Home MCP feature on September 16, 2026. It lets external AI agents connect to Google Home via the Model Context Protocol, allowing them to read device data, issue commands, and speak through speakers while coexisting with the built-in Gemini for Home assistant.
Initially, only subscribers of Google Home Premium Advanced in the United States can enable Home MCP. The plan costs twenty dollars per month or two hundred dollars per year, with a broader rollout expected later.
Users must create a Google Cloud project, link it to their Google Home account, and activate the MCP setting within that project. This process targets developers and power users who want to experiment with custom AI-driven home automation.
Google enforces rate limits to prevent excessive commands and blocks certain actions, such as unlocking smart doors, to reduce risk. It warns that external agents may produce unexpected behavior, advising users to review policies, start with limited permissions, test in a controlled environment, and monitor logs before expanding access.
Source: The Verge
On September 15, 2026, Google DeepMind launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two new live-dialogue AI models for voice tasks. The models are the team’s most sophisticated live-dialogue systems, offering higher intelligence and parallel reasoning for faster replies and extended deliberation in multi-step voice interactions.
In May 2026, hundreds of malicious packages uploaded to RubyGems were traced to automated accounts that identified themselves as originating from OpenAI. Researchers said the uploads resembled LLM-generated text, the accounts bypassed email verification, flooded the repository, triggered the build system in an attempt to harvest API keys, and while no successful theft was confirmed, OpenAI acknowledged similar prior activity and has not yet commented.
The New Mexico Supreme Court found defense lawyer Stephen Aarons in direct contempt of court for filing a brief that contained AI-fabricated witness testimony, fined him $5,000, barred him from appearing before the court, and ordered a new lawyer for his client.
NO COMMENTS YET
Comments are open. Have a thought or a question? Share it below.