Google has opened early access to Home MCP, a new integration that allows compatible AI agents to interact with Google Home smart-home environments. MCP stands for Model Context Protocol, an open standard that connects AI applications with external tools and data sources. Through Home MCP, authorized agents can discover devices, inspect current states, access historical information, and execute supported control actions.
Claude is one example of an MCP-compatible AI client that can connect to this system. This represents a significant shift from AI merely answering questions to AI interacting with physical devices in the real world. However, it is important to note that Home MCP is currently in Early Access and is not a universal feature available to every Google Home user.
What Is Google Home MCP?
Model Context Protocol (MCP) is an open standard developed by Anthropic that allows AI applications to connect with external data sources and tools. An MCP server acts as a proxy between smart home infrastructure and AI applications, exposing executable tools that allow AI clients to query real-time home telemetry and issue commands.
Google Home MCP acts as a bridge between the Google Home ecosystem and compatible AI applications. The AI client communicates with Google Home through the MCP server, which provides tools that let the AI application request information or perform supported actions. The simple conceptual flow is:
User → AI Agent → Home MCP → Google Home → Smart Device
It is important to understand that MCP itself is not an AI model. It is a protocol that provides a standardized way for AI applications to access external tools and data. The difference between an ordinary chatbot and an AI agent with tool access is significant: a chatbot generates text responses, while an agent with MCP tool access can take actions and retrieve real-world information.
How Google Home MCP Works
The architecture works through a series of steps that begin with a user instruction and end with a device action.
- User gives an instruction to an AI agent, such as “Turn off the outside lights.”
- The AI agent determines whether a Google Home tool is required to fulfill the request.
- Home MCP provides access to relevant Google Home resources and tools.
- The agent retrieves device information or state using tools like
list_home_states. - For supported actions, the agent can request the action through the MCP server using
run_home_actions. - Google Home executes the permitted command on the target device.
- The result is returned to the AI agent, which can then inform the user.
Authentication and authorization are required at every step. A random AI application does not automatically gain access to a person’s home. Users must explicitly authorize any AI client that connects to their Google Home account.
What Can AI Agents Do With Google Home MCP?
Google’s documentation describes five core capability areas that Home MCP provides.
1. Discover homes and structures
Agents can retrieve accessible homes and structures using the list_homes tool. This allows the AI to understand which home environments it has been granted access to.
2. Discover devices
Agents can identify available devices, rooms, traits, attributes, and command schemas using list_home_resources. This means the AI can learn what devices exist in the home and what commands each device supports.
3. Monitor device states
Agents can inspect current device states and connectivity information using list_home_states. This allows the AI to answer questions about whether a device is on, off, or connected.
4. Control supported devices
Agents can execute supported parameterized actions using run_home_actions. This allows the AI to turn devices on or off, adjust settings, or perform other supported operations.
5. Analyze historical events
Agents can retrieve historical state changes and event information using list_home_history. This allows the AI to answer questions about what happened in the home over a specific time period.
Examples of supported interactions include asking “How many lights do I have?”, “Is my home secured?”, “Turn off the outside lights”, or “What happened while I was away?” These are examples of supported Home MCP interactions rather than promises that every device supports every command. Device-specific capabilities can vary significantly.
Can Claude Control Google Home Devices?
Yes, Claude can interact with supported Google Home devices through the Home MCP integration when properly authorized. Google’s documentation explicitly names Claude Cowork as an MCP-compatible AI client that can connect to the Home MCP server.
However, several important qualifications apply. Authentication is required, and the user must authorize access through Google’s OAuth flow. Claude does not automatically receive access to a user’s Google Home account. The available capabilities depend on the Home MCP tools and supported device traits.
It is more accurate to say that Claude can perform supported actions on authorized devices rather than claiming it can control everything in a house. The scope of control is defined by the MCP tools Google exposes and the devices that support those tools.
Which AI Agents Work With Google Home MCP?
Google’s documentation identifies several AI clients that are officially documented as compatible with Home MCP.
- Claude Cowork: Anthropic’s MCP-compatible AI client
- Google Antigravity: Google’s agent development environment
- OpenClaw: An open-source AI agent
- Hermes: Mentioned in Google’s announcement as a supported agent
Other MCP-compatible clients may also work as the ecosystem develops, but compatibility cannot be assumed unless verified. It is important to distinguish officially documented compatibility from potential MCP compatibility. Any agent that supports calling MCP tools could theoretically connect, but Google’s documentation provides explicit guidance for the clients listed above.
What Devices Can Google Home MCP Control?
Capability depends on the devices and supported traits exposed through Google Home. The integration works with devices across the Google Home ecosystem, including Nest cameras and doorbells, Nest thermostats, and Matter-compatible devices like light bulbs.
Examples of device categories that may be supported include lights, thermostats, cameras, switches, and other compatible smart-home devices. However, not every Google Home, Nest, Matter, or third-party device supports every MCP action. Device-specific capabilities can vary based on the traits each device exposes and the command schemas it supports.
Can Google Home MCP Access Camera History?
Home MCP includes historical-analysis capabilities that allow agents to query past state changes and event logs over specific time ranges. This includes access to supported event and history information, including camera-related information where supported and authorized.
Google provides examples such as asking an agent to summarize what the kids did when they got home from school, with the agent providing a summary across cameras and showing relevant clips. Another example involves using device state history to track how many loads of laundry were done in a week.
It is important to distinguish access to supported event and history information from unrestricted camera surveillance. An AI agent cannot freely watch or download every camera recording. The access is mediated by the MCP tools Google provides and the permissions the user grants.
Google Home MCP Security and Privacy
Security and privacy protections are central to Google’s implementation of Home MCP.
Google’s documented safety protections include user authorization through OAuth authentication, access revocation that allows users to revoke an agent’s access at any time from the Google Home app or My Accounts page, rate limits that restrict how many operations an agent can perform in a given time period, and prohibited sensitive actions.
Google explicitly states that sensitive actions such as unlocking doors are prohibited through Home MCP. This is a deliberate safety measure that prevents an AI agent from performing actions that could compromise physical security.
However, Google also warns that “depending on your agent, connecting it to Home MCP can result in unexpected or even undesired behavior”. Users are advised to carefully review developer policies and terms of service before connecting an agent.
The feature expands AI’s ability to act in the physical world, so permission management becomes especially important. Users should inform other household members if they choose to connect a Google Home used by others with Home MCP, as the agent can control devices and access home data.
What Are the Risks of Giving an AI Agent Smart-Home Access?
Several practical risks are associated with giving an AI agent access to a smart home.
Unintended commands can occur if an AI agent misinterprets a natural-language request. Incorrect interpretation of ambiguous instructions could lead to actions the user did not intend. Excessive permissions granted during setup could give an agent more access than necessary. Access to home information means the agent can see device states and potentially sensitive data about home activity. Historical event data exposure could reveal patterns about when people are home or away. Shared-household privacy concerns arise when an agent has access to a home used by multiple people. A compromised AI account could allow an attacker to control home devices. Misconfigured OAuth credentials could create security vulnerabilities. Experimental and early-access behavior means the system may have bugs or unexpected limitations.
These risks should be understood as part of the trade-off for the convenience and capabilities that Home MCP provides. The risks can be managed through careful permission management, using trusted AI clients, and starting with low-risk commands.
Google Home MCP Early Access: Who Can Use It?
Home MCP is currently in Early Access and is not a general worldwide consumer rollout.
According to Google’s documentation, the requirements include:
- An active smart home setup in Google Home with connected devices
- An active Google Home Premium Advanced subscription
- Access to a Google Cloud project
- An MCP-compatible AI client such as Claude Cowork, Google Antigravity, or OpenClaw
The early access rollout began on September 16, 2026, and is initially available in English to Google Home Premium Advanced users in the United States. Google has not announced a timeline for expanding availability to other markets, including India.
The Google Home Premium Advanced subscription is priced at $20 per month or $200 per year in the United States.
How to Connect Claude to Google Home MCP
The setup process involves several steps that require some technical familiarity.
- Have a supported Google Home environment with connected devices.
- Meet the subscription requirements, including an active Google Home Premium Advanced subscription.
- Create or use an appropriate Google Cloud project in the Google Cloud console.
- Enable the Home API under APIs & Services in the Google Cloud console.
- Configure OAuth credentials by setting up the OAuth consent screen and creating an OAuth client ID.
- Add the Home MCP server to the supported AI client using the server URL and OAuth credentials.
- Complete Google’s authorization flow to grant permissions.
- Select the appropriate home structure during authorization.
- Verify available MCP tools to confirm the connection is working.
- Test with a low-risk command such as checking device state before attempting control actions.
Users should store credentials securely and never share Client IDs, Client Secrets, or OAuth tokens. Real credentials should never be exposed in documentation or public forums.
What Is the Difference Between Google Home MCP and Gemini for Home?
Home MCP and Gemini for Home serve different purposes within the Google ecosystem.
| Feature | Google Home MCP | Gemini for Home |
|---|---|---|
| Main purpose | Connect external AI agents to Google Home | Google’s own AI-powered home experience |
| AI client | External MCP-compatible clients (Claude, OpenClaw, etc.) | Google ecosystem (Gemini) |
| Protocol | Model Context Protocol | Google product experience |
| Device interaction | Supported Home MCP tools | Google Home/Gemini capabilities |
| Audience | Developers and personal AI-agent users | Google Home users |
Home MCP does not replace Gemini for Home. Gemini remains Google’s interface for interacting with Google Home through the Home app and Nest speakers. Home MCP adds an additional layer that allows third-party AI agents to interact with the home through their own interfaces.
Google Home MCP vs Traditional Smart-Home Voice Commands
The difference between traditional voice commands and agentic AI with MCP is substantial.
Traditional voice assistant: User → predefined command → smart-home action. The user speaks a specific command that the assistant recognizes and executes. The command structure is typically fixed and limited to what the voice assistant was programmed to understand.
Agentic AI + MCP: User → natural-language goal → AI reasoning → tool selection → Home MCP → device action. The user expresses a goal in natural language, and the AI agent reasons about how to achieve it, selects the appropriate tools, and executes actions through MCP.
MCP gives the AI agent a standardized way to access external tools and data. The AI has more flexibility in interpreting requests and determining which actions to take. However, this does not mean the AI has unlimited autonomous control. The MCP server defines what tools are available, and the agent can only perform actions that those tools support.
Why MCP Matters for Agentic AI
MCP represents a significant step in the evolution of AI agents from text generators to actors in the physical world.
The broader technology significance includes AI agents moving beyond chat to performing actions, tool calling as a mechanism for AI to interact with external systems, external services and APIs that AI can access through standardized interfaces, real-world actions that have physical consequences, standardized interfaces that reduce integration complexity, smart-home automation that can be more flexible and intelligent, and context-aware assistants that understand the state of the physical environment.
Connecting AI models to physical environments is fundamentally different from simply generating text. When an AI agent controls a light or thermostat, the result is a real-world change that affects people’s lives. This is why security, authorization, and safety protections are critical components of systems like Home MCP.
Real-World Examples of Google Home MCP
The capabilities of Home MCP can be understood through practical examples.
Morning routine: An authorized AI agent could check relevant home states and control supported devices to adjust lighting, temperature, or other settings based on the user’s preferences or schedule.
Leaving home: An agent could help check device states and turn off supported lights, providing a convenient way to ensure the home is in the desired state before departure.
Home monitoring: An agent could summarize supported historical events, such as what happened while the user was away or what activities occurred in specific areas of the home.
Energy awareness: An agent could inspect supported device states and help identify unusual patterns, such as lights left on for extended periods.
These examples depend on device support, permissions, and the actual AI client’s capabilities. They are illustrations of what is possible with Home MCP rather than guarantees that every setup will support every scenario.
Limitations of Google Home MCP
Home MCP has several current limitations that users should understand.
The system is in Early Access, meaning it may have bugs or incomplete features. Supported client requirements mean that not every AI agent can connect. Subscription and access requirements restrict availability to Google Home Premium Advanced users in the US. Device compatibility varies, and not every device supports every MCP action. Experimental traits may not be fully stable.
Latency is a known issue, with Google noting high latencies in current implementations. There is no universal device support, meaning some devices may not work with MCP at all. Sensitive actions restrictions prohibit certain operations like unlocking doors. Authentication complexity means setup requires technical knowledge. Automations support limitations mean agents cannot currently create automations via Home MCP.
Is Google Home MCP Safe to Use?
The answer is nuanced. Google has implemented security protections, including authorization requirements, rate limits, and restrictions on sensitive actions like unlocking doors. These protections are meaningful and reduce the risk of unintended actions.
However, connecting an AI agent to a real home introduces additional permissions and operational risks. Google itself warns that connecting an agent to Home MCP can result in “unexpected or even undesired behavior”.
Users should take several precautions:
- Review permissions carefully and grant only what is necessary.
- Use trusted AI clients from reputable developers.
- Protect credentials and never share OAuth tokens.
- Understand which devices are exposed to the agent.
- Inform other household members about the connection.
- Revoke access if it is no longer needed.
- Start with low-risk commands such as checking device states before attempting control actions.
The feature should be used with awareness of both its capabilities and its risks.
What Google Home MCP Could Mean for the Future of Smart Homes
Home MCP could enable several future developments in smart-home technology.
Proactive home assistance may become possible, where AI agents anticipate needs based on patterns and context. Natural-language automation could allow users to create complex routines using everyday language. Intelligent energy management could optimize energy usage based on occupancy, preferences, and external factors.
Contextual home monitoring could provide more meaningful summaries of home activity. Cross-device coordination could allow agents to orchestrate actions across multiple devices and systems. Personalized routines could adapt to individual preferences and schedules. AI-powered troubleshooting could help diagnose and resolve device issues.
These potential developments represent possibilities rather than guarantees. The future of Home MCP depends on how Google expands the feature, how developers build on it, and how users choose to adopt it.
Frequently Asked Questions
What is Google Home MCP?
Google Home MCP is an integration that allows compatible AI agents to interact with Google Home smart-home environments through the Model Context Protocol. Agents can discover devices, inspect states, access history, and execute supported actions.
Can Claude control Google Home devices?
Claude can interact with supported Google Home devices through Home MCP when properly authorized. Google’s documentation names Claude Cowork as a compatible client. However, authentication is required, and Claude cannot perform actions outside the supported MCP tools.
What does MCP stand for?
MCP stands for Model Context Protocol. It is an open standard developed by Anthropic that connects AI applications with external tools and data sources. MCP provides a standardized way for AI agents to access external services.
Which AI agents support Google Home MCP?
Google’s documentation names Claude Cowork, Google Antigravity, OpenClaw, and Hermes as compatible clients. Any agent that supports calling MCP tools may also work, but compatibility should be verified.
Can Google Home MCP control lights?
Yes, agents can execute supported parameterized actions on devices including lights. The run_home_actions tool allows agents to control supported devices. Not every light may support every command, depending on its traits and capabilities.
Can AI agents access Google Home history?
Yes, Home MCP includes historical-analysis capabilities. Agents can query past state changes and event logs over specific time ranges using list_home_history. This includes camera-related information where supported and authorized.
Can Google Home MCP unlock doors?
No, Google’s documentation states that sensitive actions such as unlocking doors are prohibited through Home MCP. This is a deliberate safety restriction.
Is Google Home MCP available to everyone?
No, Home MCP is currently in Early Access. It is available to Google Home Premium Advanced subscribers in the United States, in English. Google has not announced a timeline for expansion to other markets.
Is Google Home MCP free?
No, Home MCP requires a Google Home Premium Advanced subscription, priced at $20 per month or $200 per year in the US. The standard Google Home subscription tier does not include MCP access.
Is Google Home MCP safe?
Home MCP includes security protections such as authorization, rate limits, and restrictions on sensitive actions. However, Google warns that connecting an agent can result in unexpected behavior. Users should review permissions carefully and use trusted AI clients.