Google Home MCP Explained: How to Let Claude Control Your Smart Home, and Whether the $20 Plan Is Worth It

What's in this guide

• The short answer
• What MCP actually is, in plain English
• What Home MCP lets your AI do (and what it refuses to do)
• What you need before you start
• Which AI assistants work with it right now
• What happened when I connected Claude to my house
• Is the $20 Advanced plan worth it for this?
• The privacy and safety questions nobody is asking
• The verdict

On September 16, 2026, Google quietly did something the smart home world has been waiting on for a while: it opened the doors of Google Home to outside AI assistants. The feature is called Home MCP, and it means Claude, Google's own Antigravity, OpenClaw and any other assistant that speaks the same protocol can now look at your Nest cameras, read your thermostat, flip your Matter light bulbs and dig through your home's activity history. You just ask in normal English.

That sounds like magic and also slightly alarming, so I spent two evenings setting it up and living with it. Here is what MCP is, what Home MCP can and cannot do, what it costs, and whether a regular person should bother.

The short answer

Home MCP is an early access bridge that lets an AI assistant you already use control and query the devices in your Google Home. It works, it is genuinely useful for camera summaries and "what happened while I was out" questions, and it is more capable than the Gemini built into the Google Home app for anything that involves reasoning across several devices or days.

But it is gated behind the $20 a month Google Home Premium Advanced plan, it is US only for now, and the setup involves creating a Google Cloud project, which is a step most households have never taken. If you already pay for Advanced and already use Claude or another compatible assistant, set it up this weekend. If you do not, it is not yet worth $240 a year on its own.

What MCP actually is, in plain English

MCP stands for Model Context Protocol. Strip the jargon and it is a standard way for an AI assistant to plug into outside tools. Think of it as USB for AI: one connector shape, so every assistant can talk to every service without each pair needing a custom cable.

Anthropic, the company behind Claude, open-sourced MCP in November 2024. By December 2025 it had grown so quickly that Anthropic handed it to a neutral body, the Agentic AI Foundation under the Linux Foundation, co-founded with OpenAI and Block and backed by Google, Microsoft and AWS. That is the important bit for you as a consumer. MCP is not a Claude thing or a Google thing. It is a shared plug, which is exactly why Google could build one Home MCP server and have several different assistants connect to it on day one.

Until now MCP lived almost entirely in developer land: connecting AI coding tools to GitHub, databases and company wikis. Home MCP is one of the first times it has shown up in a product that ordinary people own.

What Home MCP lets your AI do (and what it refuses to do)

Google's server exposes five tools to your assistant, and once you know them you can predict what any prompt will do:

• List homes. Finds the home structures on your account.
• List resources. Reads every device, room, and the commands each device supports.
• List states. Checks live status: is the front door camera online, what is the thermostat set to, which lights are on.
• Run actions. Sends commands: turn off the outside lights, set the living room to 70 degrees.
• List history. Pulls past events and state changes over a time range you specify.

That last one is where it gets interesting. Google's own examples include asking what the kids did after school and getting a summary across every camera with clips, or asking how many loads of laundry ran this week based on a smart plug's on and off history. The built-in Gemini in the Google Home app can search your video history, but it cannot chain that into "compare this week to last week and tell me if the delivery driver came earlier."

Now the refusals. Home MCP prohibits what Google calls sensitive actions, and unlocking doors is the example it gives. There are rate limits too. Creating and editing automations is not supported yet; Google says that is coming in a future release. And familiar face data (the names your Nest cameras attach to people) needs a separate, explicit consent link before your assistant can see it.

What you need before you start

Here is the checklist, taken from Google's Home MCP user guide and confirmed against the Google Store pricing page as of September 18, 2026:

• A Google Home with devices already set up in the Google Home app.
• A Google Home Premium Advanced subscription. That is the top tier: $20 a month or $200 a year. The $10 Standard tier does not qualify. New households get a 30 day trial, and the old Nest Aware Plus plan was converted to Advanced automatically.
• A US location. Google told TechCrunch it would not comment on when other countries or tiers get access.
• A Google Cloud project. It is free to create, but you will need to enable the Home API, set up an OAuth consent screen and generate a client ID and secret. Google's guide walks through it, and you can paste a ready-made prompt into your assistant to have it do the client-side configuration for you.
• An assistant that supports MCP with sign-in. Google's guide gives step by step instructions for Antigravity, Claude and OpenClaw.

Which AI assistants work with it right now

Google's announcement names Antigravity, Claude, Hermes and OpenClaw, and TechCrunch's report adds ChatGPT to the list of MCP capable agents. In practice, the ones with an official setup walkthrough on Google's page are these three:

Claude is the easiest path for most people. You add Home MCP as a custom connector on the claude.ai connectors page, paste in your client ID and secret, approve the Google sign-in, then toggle it on inside a Cowork chat. If you are still deciding between assistants, my Claude vs ChatGPT vs Gemini comparison covers the plans, and the Claude Fable 5.1 explainer covers which tier you actually need.

Google Antigravity is Google's own agent, aimed at developers, with a settings file you edit by hand. It works, but it is the least friendly setup of the three.

OpenClaw is the option for people who run their assistant locally and want the connection to live on their own machine. It uses a config patch and a terminal login command. If you have followed my local AI guide, this will feel familiar.

ChatGPT supports custom MCP connectors through its developer mode on paid plans, but Google has not published ChatGPT specific steps and the redirect URL you need is not listed in its guide. I would wait for an official walkthrough rather than guess.

What happened when I connected Claude to my house

The Google Cloud part took me about 25 minutes, most of it reading OAuth consent screen forms that were clearly written for app developers, not people who want to ask about their doorbell. Once I had the client ID and secret, the Claude side took under five minutes. I pasted Google's setup prompt, approved the sign-in, picked my home, and it was connected.

First prompt: "How many lights do I have in my house?" It came back with 14, grouped by room, which matched the Google Home app exactly. Second: "Is my home secured?" It reported every camera online, the front door locked, and the garage closed. Third: "Turn off all the outside lights." Both porch lights and the driveway floodlight went off within a few seconds, and Claude asked me to confirm before it sent the command.

The prompt that sold me was "What happened while I was out today?" I got a clean paragraph: a package delivery at 11:40, the dog walker at 1:15, a person lingering by the mailbox at 3:05 that turned out to be a neighbor, and a note that the thermostat had been bumped to 74 by someone at home.

Two negatives. Responses regularly took eight to twelve seconds, which Google lists as a known latency issue it is working on. And when I asked it to unlock the front door it correctly refused, but without explaining why, so you could easily think the connection had broken.

Is the $20 Advanced plan worth it for this?

The honest answer depends on where you are starting from.

You already pay for Advanced. Set it up. The marginal cost is zero and the camera history summaries alone justify an hour of setup.

You pay for Standard ($10). The jump to Advanced is $120 a year. If you have two or more Nest cameras or a doorbell, the Advanced features you also get (60 days of event history, 10 days of continuous recording, searchable video, daily summaries) are arguably worth that on their own, and Home MCP becomes a bonus. If you only have a thermostat and some bulbs, stay on Standard for now.

You pay nothing today. Do not subscribe just for this. Use the 30 day trial if you are a new household, try Home MCP during the trial, and decide with real experience. Early access features have a habit of moving to lower tiers once they leave beta, and Google has said nothing either way.

You have no Nest cameras at all. The value drops sharply. Without cameras, Home MCP is mostly a fancier way to toggle lights and read a thermostat, which Gemini in the Home app already does for free.

The privacy and safety questions nobody is asking

The launch coverage focused on what you can do. I want to spend a minute on what you are handing over, because I just wrote about smart TVs quietly sending screenshots home and the same instincts apply.

When you connect an assistant to Home MCP, that assistant's provider can now receive descriptions of your camera events, device states and history whenever you ask a question. For Claude that means Anthropic's servers see the text of those events. Google's own warning page says as much and adds that you should tell everyone else in your household, because your agent can now see and control shared devices. Google also suggests creating a separate test home with a few devices if you want to experiment without exposing the whole house.

Two practical steps. First, check your assistant's data training settings. I covered how to stop ChatGPT, Claude and Gemini training on your chats recently, and it matters more once your chats contain your doorbell history. Second, remember that access can be revoked at any time from the Google Home app or your Google Account's third party access page. Do that the moment you stop using it.

The bigger risk is the one Google flags itself: depending on your agent, connecting it can result in unexpected or undesired behavior. Give any AI that can switch things in your home a confirm-before-acting rule, and do not let it run unattended tasks until you have watched it work for a while. On September 15, Spain's data protection agency logged what it called the first data breach carried out end to end by an AI agent. Home MCP is not that, but it is the same category of tool.

The verdict

Home MCP is the first consumer feature that made MCP click for me. It is a shared plug that lets the assistant I already talk to reach into the house I already own. The camera summaries are excellent, the device control works, and the refusal to unlock doors is the right default.

The catch is entirely commercial. Early access, US only, and locked to the $20 tier. If you are already there, this is the best reason yet to have paid for Advanced. If you are not, wait. Google has every incentive to push this to more plans and more countries once it leaves early access, and by then automations support should have landed, which is when Home MCP goes from neat trick to something I would actually build my house around.