Google opened early access to Home MCP on September 16. Meta filed smart home as a built-in connector category inside Muse. Sonos and Home Assistant signed on. Amazon too.
Six weeks. That is the entire elapsed time between MCP being a protocol proposal and being surrounded by every major smart home platform except one.
Apple's Command Center — the $350, seven-inch hub — slipped to late 2026 or early 2027. Siri AI shipped on the phone on September 14. Conversational HomePod control did not.
The hardware delay is the headline. It is not the story.
The story is that a protocol layer is being standardized without the largest installed base of premium home devices. Where code becomes law in the digital frontier, whoever writes the interface writes the tax.
MCP, the Model Context Protocol, is not a model architecture. It is not a chip. It is an interop layer — the thing that decides how an AI agent reads a device state, requests an action, and receives confirmation. It sits one level above Matter and Thread. Matter answers whether these devices can connect. MCP answers whether an agent can understand and command them.
I have watched this movie before. In 2017, while finishing my computer science thesis, I spent forty hours a week auditing ERC-20 contracts across more than fifty ICOs. I found reentrancy vulnerabilities in three major raises. What I learned was not about tokenomics. It was about how standards win. ERC-20 was not the most elegant interface. It was the interface the most wallets happened to implement. Distribution beat design.
The same logic is executing now on home devices. MCP's six-week adoption curve is not a technical validation — it is a network-effect flywheel spinning up in public.
The mechanics matter. A cloud-side agent needs context. To control a light, a lock, a thermostat, it needs state, intent, and history. That is data leaving the device. Apple's privacy-first, on-device architecture is theoretically superior — I spent six months in 2022 optimizing zk-SNARK circuits for a Layer 2, cutting proof generation time by 15%, and I know exactly what on-device verification buys you. It buys a smaller attack surface and a cleaner compliance story.
It also buys a structural conflict. An agent running on Google's cloud cannot reason about your home if your home will not talk to it. Apple's privacy architecture is correct in theory and friction-locked against the subscription economics the rest of the industry has already agreed on.
And they have agreed. Google Home Premium Advanced, Meta Muse Power, Alexa+ — all converge near $20 per month. That number is the tell. When three competitors independently land on the same price point within weeks, they are not guessing. They are copying a unit-economic model that already works.
In a subscription regime, hardware is a customer acquisition cost. A $350 hub with no attached recurring agent service looks less like a product and more like a subsidy Apple refuses to pay. Margin migrates to the orchestration layer — the same migration I documented when Uniswap V2 automated market maker mechanics turned liquidity provision from a human activity into a protocol primitive. Value does not sit where the asset is held. It sits where the routing happens.
Here is the contrarian read, and it is the one I would defend in a technical review.
MCP is not a standard yet. It is a press cycle with good timing. Six weeks of announcements is not governance. There is no published permission model, no minimum-privilege specification, no audit-log format, no revocation path. Ask the simple question: when a prompt-injected agent unlocks a front door at 3 a.m., whose logs prove what happened?
Nobody has answered that. Not Google. Not Meta. Not Amazon.
The architecture of trust, stripped to its bones, is authentication plus authorization plus attribution. MCP has shipped the connector surface and none of the three. That is not a standard. That is a demo.
I have seen this failure mode at close range. Every interoperability layer I have audited follows the same arc: connectors first, permissions later, incident reports last. The security substrate arrives only after the first physical-world harm.
So the consensus trade — Apple loses the smart home — may be directionally right and tactically premature. "Support" is cheap to announce. Deep integration is expensive to build. Watch whether Google's Home MCP early access converts to general availability on a published timeline. Watch whether Sonos and Home Assistant are shipping real connectors or logos on a slide. The gap between announced and integrated is where most protocols die quietly.
Apple also is not dead. It has cash, a chip stack, and a captive premium user base. It can buy an agent team or an MCP tooling vendor tomorrow. But the migration cost curve is real. Once a household has wired lights, locks, cameras, and a $20 monthly agent subscription into one ecosystem, switching means re-teaching an ambient assistant how you live. That friction is the moat. It is not Apple's moat right now.
Navigating the storm with empirical precision means watching three signals, not three headlines.
One signal: whether MCP governance formalizes before Q4 2026 — a named body, a permission spec, an auditable revocation mechanism. Without it, the standard is a liability transfer.
Another: whether Apple announces MCP compatibility or a home agent subscription at WWDC. Silence there is a strategic decision, not a scheduling one.
The third: whether the $20 tier retains users past month three. Subscription smart home has no public churn data. Everything above assumes retention that has not been demonstrated.
The tradeable layer here is not the assistant. It is the same layer crypto infrastructure has spent eight years bleeding over: permission management, cross-protocol bridging, audit trails. Agent identity. Scoped device credentials. Verifiable logs of who let the agent do what.
Clarity emerges from the chaos of verification. The agents will ship. The question is whether anything verifies what they did.