Google Home opens its MCP server to any AI agent

Google Home just opened its MCP server, which means any MCP-compatible AI agent — Claude, ChatGPT, OpenClaw, Google's own Antigravity — can now read your home's event history, control your devices, and build its own dashboards for them. The rollout landed as early access this week for Google Home Premium Advanced subscribers in the US, and the setup guide is already live in Google's Home developer docs. Google would not say when or whether the integration reaches other subscription tiers or markets.

What the agent actually gets is broader than turning lights on. The Home MCP server exposes tools that let a model inspect home structure, query real-time device state, pull historical events, and issue control commands. In practice that means cross-camera analysis — asking your agent what your kid did when they got home from school — or digging through device state history to count how many loads of laundry ran last week. Devices can also answer back through the system: an agent that finishes a long task can push an audio message out of a Google Home speaker, and it can assemble a custom dashboard to run the house from.

Setup is not plug-and-play, and that is the interesting part. You create your own Google Cloud project, enable the Home API, configure an OAuth client, then hand the MCP configuration to your agent and walk through a consent flow that grants it access to a specific home structure. Familiar-faces data — the identity layer behind Nest camera alerts — requires a second, separate consent. Access is per-home and revocable, which is the right shape for a permission model; whether agents honor it cleanly once they hold the keys is the open question.

That question is not hypothetical. MCP has been the fastest-spreading agent standard of the year, and the hard problems have been identity and provenance: proving which agent asked for what, on whose behalf, with what scopes — the plumbing problem we wrote about in August. Google is now pushing that untested permission layer into a device class where a bad tool call opens a front door lock instead of deleting a draft. The upside is real: a system that reasons across cameras, thermostats and appliances is meaningfully more useful than one that answers "I'm sorry, I can't do that." The cost is that your home's activity log becomes an input to a model whose provider you chose, with its own retention and training policies — none of which Google controls.

Worth knowing if you have not met the protocol yet: our explainer on what MCP actually is covers the client-server split in plain English, and it is the right primer before you hand a cloud project's OAuth credentials to an agent.

What to watch: whether Google widens Home MCP beyond Premium Advanced, and whether it publishes audit tooling for what agents did in your house — a scoped, logged, revocable trail is the difference between an assistant and an unknown roommate.

Would you give an AI agent write access to your smart home, or is the read-only version the only sane default? Tell us in the comments.

Byteification turns Qwen3 and Llama 3 into byte-level models

The tokenizer is the last piece of the language-model pipeline that was never learned end to end — and a Nature paper out this week shows it can be swapped out for less than one percent of a pretraining budget. Researchers have retrofitted four open subword language models into true byte-level models using less than 1% of a typical pretraining budget, and the result — published 7 October in Nature — closes the performance gap that has kept byte-level models a niche curiosity for nearly a decad

OpenAI's dots go mobile and can start your Codex work

One product update and two benchmark papers this hour, all circling the same question: how much of the work can you actually hand over, and what happens to the part you didn't? OpenAI's dots — the always-on agents that get their own cloud computer — can now be created and configured entirely from the ChatGPT app on iOS and Android, and a dot can start Codex work or pick up an existing Codex thread on your behalf. The October 9 release, billed as the first of a fresh run of updates for dots, mo

Coding agents as a hospital: Cockroach's five-month results

The agentic-coding debate keeps being framed as speed versus trust. Cockroach Labs just published the most detailed trust-first ledger yet — five months of numbers, including the parts that didn't work. Cockroach Labs ran its codebase like a teaching hospital for five months and says the pipeline merged well over a million lines of code with seven total reverts. In the setup the company calls MOLT Sinai, GitHub issues are patients, merging is discharge, and the humans in charge are Chiefs of M

Today in AI — October 10, 2026

A day where the boring machinery took center stage: proof checkers, order books, warehouses and the humble text message — plus fresh arguments about who gets to declare any of it safe. Models & Research * Mathematicians get a reliability primer for the Lean Theorem Prover. A guest post by Thomas Hales on Terence Tao's blog walks through what Lean actually guarantees — and devotes itself to the "Summer of Soundness Bugs," the string of kernel bugs found this summer that let false proofs thr

Read more

Byteification turns Qwen3 and Llama 3 into byte-level models

OpenAI's dots go mobile and can start your Codex work

Coding agents as a hospital: Cockroach's five-month results

Today in AI — October 10, 2026