On August 27, 2026, Anthropic unveiled its Model Hardware Standard (MHS), a software specification that lets AI assistants like Claude interact directly with physical hardware — factory robots, scientific lab instruments, and manufacturing equipment — using structured, machine-readable descriptions instead of paper manuals or tribal knowledge.
The protocol is built on top of MCP (Model Context Protocol), the open standard Anthropic introduced in 2024 that has since become the industry baseline for connecting AI agents to software systems. MHS extends that idea into the physical world: a robot-arm vendor, for instance, can encode in an MHS file exactly how an AI should manipulate the arm safely — capping speed, restricting joint angles, enforcing load limits — and Claude reads that specification rather than guessing. The result is structured safety, baked in at the protocol layer.
The standard is deliberately model-agnostic. It isn't locked to Claude; any AI model that can parse an MHS file can use it. That's a smart play: by designing for interoperability rather than lock-in, Anthropic is positioning MHS as infrastructure rather than a product feature.
Early partners already testing the research preview include AWS, Danaher, Hugging Face, and Raspberry Pi — a mix of enterprise data-center, scientific instruments, AI tooling, and hobbyist hardware that signals how broad Anthropic's ambitions are for the standard. A waitlist is now open to additional developers in scientific, robotics, and manufacturing fields.
This is Anthropic's first direct move into physical-world AI. The implications are significant: MHS could make it far easier to deploy AI agents in laboratories, hospitals, and factories without requiring custom integrations for every piece of equipment. Instead of each manufacturer shipping bespoke AI connectors, they ship a single MHS descriptor and let any capable AI model pick it up.
CNBC noted that Anthropic plans to fully open-source MHS in the future, once the research preview has generated enough real-world feedback to stabilize the specification. For now, the waitlist is how you get access.
The timing tracks with a broader industry shift. As AI models grow capable of managing long-running, multi-step tasks, "physical AI" — agents that can actually do things in the real world rather than just in software — has become the next frontier. MHS positions Anthropic alongside robotics-focused AI efforts from Google DeepMind and others, but with a standards-first, open-protocol approach that could drive adoption across vendors who'd otherwise compete on proprietary integrations.
MCP surpassing 400 million monthly SDK downloads earlier this year — a 4× increase — shows that when Anthropic builds open protocols, the industry adopts them. If MHS follows a similar trajectory, the physical world just got a lot more accessible to AI agents.