Google 将允许任何 AI Agent 运行你的智能家居

Google 正在向 AI Agent 开放其智能家居平台，使 Claude、Open Claw 等工具能够通过标准化的 Model Context Protocol，访问和控制你的已连接设备，并分析你家的数据。

Google Home MCP 是一项新集成，可让第三方 AI Agent 控制和监测你的智能家居，并代表你采取行动。Google Home 与 Nest 集团产品经理 Taylor Lehman 在一篇博客文章中表示，它“允许任何支持 MCP 的 AI Agent——包括 Google Antigravity、Claude、Hermes 或 Open Claw——安全地使用 Google Home 生态系统中的所有设备和事件历史记录”。

Lehman 表示，Home MCP 集成将把你选择的 AI Agent 连接至现实世界事件，解锁一系列新能力。例如，它可以进行跨摄像头分析——你可以询问 Agent，你的孩子放学回家后做了什么；也可以利用设备状态历史，统计你上周洗了多少次衣服，或灯被开着多久。它还允许 Agent 通过语音与你交互，例如在任务完成后，通过 Google Home 扬声器向你发送音频消息。你还可以让 Agent 创建一个用于控制 Google Home 的自定义仪表盘。

Google Home 智能音箱仍将由 Gemini for Home 控制，但启用 MCP 后，第三方 AI Agent 将能够通过这些音箱向你发送语音消息。 图片：Jennifer Patison Tuohy / The Verge

这并不会取代 Google 的新 Gemini for Home 助手。在Google Assistant 退出历史舞台后，Gemini for Home 仍是 Google 通过 Home 应用和 Google Nest 智能音箱与 Google Home 交互的界面。MCP 集成增加的是另一层控制能力，让第三方 AI Agent 能够通过自己的界面，直接与连接至 Google Home 的设备交互。

如果本周没有被“AI 将如何毁灭全人类”的新闻标题主导，我会对此感到非常兴奋。

在发布之初，该功能仅限美国的Google Home Premium Advanced 用户使用，价格为每月 20 美元或每年 200 美元，访问权限将在未来几周内逐步推出。设置过程需要创建一个 Google Cloud 项目，并将其配置为使用 Home MCP。

让 AI Agent 访问智能门锁、恒温器等现实世界设备，以及 HVAC 系统和家电等核心基础设施，显然会引发安全、隐私和人身安全方面的问题。Google 表示，Home MCP 会实施速率限制和安全保护措施；例如，它不会允许 Agent 解锁房门。不过，Lehman 表示，取决于你使用的 Agent，“将其连接至 Home MCP 可能导致意外甚至不受欢迎的行为”。他建议用户审阅 Google 的开发者政策和服务条款。

开源智能家居平台 Home Assistant 已实施类似的 MCP 集成，展示了当你让一个 Agent 访问智能家居系统时能够实现什么。去年，我测试了用 Claude 对智能家居进行 vibe coding。它能轻松处理许多大多数界面——以及用户本身——都难以完成的复杂任务，包括故障排查、创建自动化、设计仪表盘和配置更高级的设置，且全部通过自然语言完成。

这里最大的变化，并不是 Google Home 现在可以通过你的 Claude 应用打开或关闭灯；而是 Claude——或者你选择的任何 Agent——能够访问你家底层的数据和控制层。这是朝着构建智能家居的完整家庭计算机迈出的一步：它能理解家中正在发生什么，根据历史进行推理，并代表你采取行动。如果过去一周没有被“AI 将如何杀死所有人类”的新闻标题所主导，我会对此感到非常兴奋。

一个能够排查故障、提出改进建议、并识别全屋模式的系统，比只会说“抱歉，我无法做到”的系统更有用。

尽管如此，让 AI Agent 访问你的智能家居确实存在实际优势。它可以跨越家庭产生的数据进行推理，并以传统智能家居系统通常无法实现的方式采取行动，让智能家居从命令与控制模式转向更具情境感知和主动性的模式，也更接近真正的智能。一个可以排查设备问题、为自动化提出改进建议、追踪并识别全屋模式的系统，远比一个只会说“抱歉，我无法做到”的系统有用。借助 Home MCP 集成，Google 正在将这些能力引入 Google Home。

Claude 使用 MCP 在 Home Assistant 中为我创建了这个仪表盘。启用 Google Home MCP 后，它应该也能为我的 Google Home 完成类似工作。 截图：Jennifer Pattison Tuohy / The Verge

尽管 Claude 等面向消费者的工具会对智能家居爱好者很有用，但更大的图景在于 Google 面向开发者的布局。过去几年中，Google 越来越将自己定位为智能家居的基础设施层。该公司在 2024 年推出了智能家居 API 访问能力，最近又将 Gemini for Home 打造为全栈 AI 产品。这一切都指向更偏向 B2B2C 的战略：Google 可以提供 AI 和智能家居基础设施，而其他公司则在其之上构建面向消费者的服务和产品。

随着本周的公告，Google 还为开发者打开了一条路径：他们可以使用其具备 Agent 能力的 Antigravity 编程工具，构建能够调用 Google Home 设备数据和控制层的定制智能家居 AI Agent。不过，即使开发者使用的是其他 Agent，只要他们采用 Google Home，对 Google 来说依然是胜利。通过这种方式，Google 将成为智能家居的基础设施层，就像 AWS 成为互联网与软件企业的基础设施一样。

智能家居开发者是否会接受 Google 的这些承诺，仍是一个重大问题。鉴于该公司多年来推出又放弃的智能家居平台数量——Android @ Home、Weave、Project Brillo、Works with Nest、Google Assistant——开发者必须判断，Google 是否终于确定了一个打算长期坚持的智能家居平台。

Google is opening up its smart home to AI agents, letting tools like Claude and Open Claw access and control your connected devices and analyze your home’s data using the standardized Model Context Protocol.

Google Home MCP is a new integration that lets third-party AI agents control and monitor your smart home and act on your behalf. It “allows any AI agents that support MCP, including Google Antigravity, Claude, Hermes or Open Claw, to securely work with all of the devices and event history in your Google Home ecosystem,” Taylor Lehman, group product manager at Google Home & Nest, said in a blog post.

According to Lehman, Home MCP integration will connect your AI agent of choice to real-world events, enabling a host of new capabilities. From things like cross-camera analysis — so you could ask your agent what your kid did when they got home from school — to using device state history to figure out how many loads of laundry you did last week or how long the lights were left on. It will also allow your agent to interact with you via voice, for example, by sending an audio message over your Google Home speaker when it has finished a task. You can also have your agent create a custom dashboard to control your Google Home.

Google Home smart speakers will still be controlled by Gemini for Home, but with MCP enabled third-party AI agents will be able to send voice messages to you through them. Photo: Jennifer Patison Tuohy / The Verge

This doesn’t replace Google’s new Gemini for Home assistant, which remains Google’s interface for interacting with Google Home through the Home app and Google Nest smart speakers, following the demise of Google Assistant. Instead, MCP integration adds another layer of control, allowing third-party AI agents to interact directly with devices connected to Google Home through their own interfaces.

If this week wasn’t dominated by headlines about how AI will kill us all, I’d be really excited about this

At launch, availability is limited to Google Home Premium Advanced users in the US ($20/month or $200/year), with access rolling out in the coming weeks. Setup requires creating a Google Cloud project and configuring it to use the Home MCP.

Giving an AI agent access to real-world devices like smart locks, thermostats, and core infrastructure like HVAC systems and appliances raises obvious questions about security, privacy, and safety. Google says Home MCP enforces rate limits and safety protections; for example, it won’t allow an agent to unlock doors. However, depending on your agent, “connecting it to Home MCP can result in unexpected or even undesired behavior,” said Lehman, who recommends reviewing Google’s developer policies and terms of service.

Home Assistant, an open-source smart home platform, has implemented a similar MCP integration that shows what you can do when you give an agent access to your smart home system. I tested this last year to use Claude to vibe-code my smart home, and it easily handled some of the more complex tasks that most interfaces — and users, for that matter — struggle with, including troubleshooting, creating automations, designing a dashboard, and setting up more advanced configurations, all using natural language.

The biggest change here isn’t that Google Home can now turn your lights on and off through your Claude app; it’s that Claude — or your agent of choice — has access to the underlying data and control layer of your home. It’s a step toward building a whole home computer for your smart home that can understand what’s happening in it, reason across its history, and take action on your behalf. And if this last week wasn’t dominated by headlines about how AI will kill us all, I’d be really excited.

A system that can troubleshoot, suggest improvements, and identify patterns across your home is more useful than one that says “I’m sorry, I can’t do that”

Still, there are real advantages to giving an AI agent access to your smart home. It can reason across the data your home generates and act on it in ways traditional smart home systems generally can’t, moving the smart home from command and control toward something more contextual and proactive, closer to actual intelligence. A system that can troubleshoot a device, suggest improvements to automations, and track and identify patterns across your home is much more useful than one that just says “I’m sorry, I can’t do that.” With Home MCP integration, Google is bringing these capabilities into Google Home.

Claude created this dashboard for me in Home Assistant using MCP. With Google Home MCP enabled, it should be able to do something similar for my Google Home. Screenshots by Jennifer Pattison Tuohy / The Verge

While consumer-facing tools like Claude will be useful for smart home tinkerers, the bigger picture here is Google’s developer play. Over the last couple of years, Google has increasingly positioned itself as the smart home’s infrastructure layer. It launched API access to its smart home in 2024 and more recently made Gemini for Home a full-stack AI offering. All of this points toward a more B2B2C strategy: Google can provide the AI and smart home infrastructure while other companies build consumer-facing services and products on top.

With this week’s announcement, Google has also opened a path for developers to use its agentic Antigravity coding tool to build custom AI agents for the smart home that tap into Google Home device data and control layer. But even if developers use a different agent, as long as they’re using Google Home, it’s a win for Google. This way, Google becomes the smart home infrastructure layer, in much the same way AWS became infrastructure for internet and software businesses.

Whether smart home developers will buy into Google’s promises here is a big question. Based on the number of smart home platforms the company has launched and let die over the years — Android @ Home, Weave, Project Brillo, Works with Nest, Google Assistant — developers will have to decide whether Google has finally settled on a smart home platform it intends to stick with.

Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.

• Jennifer Pattison Tuohy

• AI

• Google

• News

• Smart Home

• Tech