Google Home Adds MCP Support, Integrates with ChatGPT and Others

On September 16, 2026, Google announced support for Model Context Protocol as a mechanism for operating Google Home from external AI agents. According to reporting by Chandra Steele of Android Police, all agents that support MCP calls are covered. Specifically, ChatGPT, Claude, Hermes, OpenClaw, and others were named. Agents can operate devices, analyze summaries, and create dedicated control screens. Google Home speakers can also be used as voice output for agents.

The announcement was made in a blog post by Taylor Lehman, product lead for Google Home and Nest. He described the new support as follows.

just the next step in unlocking agentic smart homes.

The remark conveys that it is just the next step in unlocking agentic smart homes. Google has chosen to open control of home equipment to external intelligence rather than limiting it to specific conversational assistants. The move is attracting attention as an attempt to connect home infrastructure with other companies’ language models.

Expanded Control Through MCP Support

The expanded control enabled by MCP support goes beyond simply replacing voice commands. With the user’s permission, agents obtain the status of devices in the home and execute operations. In addition to controlling lighting and home appliances, they also handle summarization and analysis of logs obtained from devices. This includes the ability to automatically assemble customized control screens for each user. The expected operations involve situational judgment, rather than fixed responses.

Handling of voice output is also changing. Agents can use Google Home speakers as their own voice. Users can receive responses from external models through speakers in specific rooms. This creates a setup in which Google’s assistant functions do not monopolize conversational touchpoints in the home. The difference in audio performance noted in New Google Home Speaker Falls Short of 6-Year-Old Nest Audio in Sound Quality is also likely to affect how uses are sorted.

Supported Devices and the Overall Ecosystem

The supported devices and the overall ecosystem carry over Google Home’s existing assets as they are. Covered devices include Nest devices, Matter-standard devices, and devices labeled “Works with Google Home.” Even if communication standards or manufacturers differ, devices registered with Google Home become operable targets. The background is the increased interoperability between devices driven by the spread of Matter. It is designed to treat the entire home as a single target of operation.

MCP is a common connection procedure for models to use external context and tools. By providing an MCP endpoint on the Google Home side, the need for individual integration development is eliminated. In principle, any agent that supports the procedure can connect. It is a structure that loosens lock-in by specific companies and encourages interconnection. The home-device platform can be seen as moving closer to an open connection layer.

Setup Steps and Paid Plan Requirements

The setup steps and paid plan requirements are currently limited in scope. Use requires U.S. residency and a $20-per-month Google Home Premium Advanced subscription. Google says it is rolling out gradually, and expects deployment to eligible users to take about one to two weeks. It is in an early-access stage, not a free offering for the general public. This can be seen as an approach of expanding the audience while repeating verification.

Setup requires creating a Google Cloud project. Users prepare a project and pass the MCP connection information to the agent to instruct setup. The agent then requests authentication and authorization. Once approval is complete, the agent connects to the home devices. Detailed instructions are published in the Google Home Developer Center.

The fact that developer-oriented procedures remain is important. It is not a method that general users can complete solely through on-screen operations. As with the testing infrastructure seen in GNOME OS Test Center Inspired by Apple TestFlight, experimentation by developers comes first. Users are required to understand how connection information is handled and the scope of permissions. For the time being, use is expected to center on technically knowledgeable users.

Safety Measures and Specific Prohibited Operations

Safety measures and specific prohibited operations have also been disclosed. In Home MCP, certain sensitive operations, such as unlocking locks, are prohibited. Because it handles indoor and outdoor cameras and audio devices as well as heating and cooling equipment, safeguards are essential. Lehman noted that depending on the agent used, unexpected or undesirable behavior may occur. Balancing openness and safety remains a challenge.

The structure of granting home operation authority to external models complicates where risks lie. Errors in a model’s reasoning or misinterpretation of instructions lead directly to physical actions. A lighting malfunction differs in severity from a lock or air-conditioning malfunction. Setting prohibited operations alone cannot prevent every situation. It is necessary to check the scope of log retention and whether data is sent to third parties.

This is not unrelated to the debate over misuse of models. As shown by Anthropic to Release Ultra-Risky AI ‘Claude Mythos’ on a Limited Basis, management of powerful models is becoming stricter. When combined with home operation, the impact is not limited to information leaks. Minimization of permissions and visualization of operation history are required. Development of control measures on the user side is being questioned in parallel.

Google’s Aims and Changing Competitive Landscape

Google’s aims and the changing competitive landscape relate to the battle for leadership of home platforms. Amazon’s Alexa and Apple’s HomeKit have centered on their own voice assistants. Google seeks differentiation by bringing in external models. It is a strategy of not limiting itself to its own models and leaving the choice to users. By positioning itself as the provider of the home platform, it aims to maintain points of contact for use.

Support for MCP also leads to bringing in developers. Because there is no need to build individual dedicated integrations, the burden of support is light. Emerging players such as Hermes and OpenClaw can also participate easily. Competition on the model side could spill over into expanded home functions. This is a structure in which opening the platform promotes ecosystem growth.

The early offering in the higher-priced paid tier also has meaning. Operational issues can be identified among the $20-per-month user base. Measured values such as audio quality, response latency, and frequency of erroneous operations will be collected. The insights gained can be applied to expansion to lower tiers and specification revisions. This can be evaluated as a method that combines monetization and validation.

Editorial Opinion

On short-term impact. Over the next three to six months, trials are expected to proceed centered on paid subscribers in the United States. Developers will compete on the accuracy of automatic screen generation and summary analysis. Home device manufacturers will likely rediscover the value of Matter compatibility and Google Home registration. Reports of erroneous operations and delays will accumulate, and the list of prohibited operations is expected to be revised.

On the long-term perspective. On a one- to three-year horizon, home platforms could become independent of specific conversational assistants. Users are expected to choose to keep using the same home equipment while switching models. Management of home operation history and life logs will likely become a new axis of competition. Institutional design to define the division of responsibility between models and platforms will be needed.

Questions from the editorial team. If unexpected behavior occurs, where does responsibility lie — with the user, the model provider, or the platform provider. Is prohibiting unlocking alone enough to ensure residents’ safety. Are means to detect hijacking of voice output or external transmission of operation history sufficient. As a premise for the spread of open agentic smart homes, establishing a framework for verification is required.

References

• “Google Home opens the door to ChatGPT, Claude, and other AI agents”, by Chandra Steele — Android Police, 2026-09-16T21:49:00.000Z (ARR)
• Source URL: https://www.androidpolice.com/google-home-opens-the-door-to-chatgpt-claude-and-other-ai-agents/