Model Hardware Standard
What it is
The Model Hardware Standard (MHS) is Anthropic's new interface specification for letting AI agents discover and safely operate physical equipment — lab instruments, robotic arms, manufacturing machinery — instead of just software tools and APIs. It targets any device with a programmable interface, is explicitly model-agnostic (not limited to Claude), and bakes operational safety constraints (speed limits, angle restrictions) directly into the standard rather than leaving them to be prompted around.
Why it's emerging now
Agent tool-use has mostly meant calling software APIs; wiring an agent to a physical device has stayed bespoke, slow integration work — Anthropic cites cutting that from weeks of engineering to hours or minutes. As agentic systems move from chat and coding into science and manufacturing (drug discovery, quantum hardware calibration), the industry needs a shared way for models to discover and control machines safely, the same gap MCP closed for software tools a couple of years earlier.
How it relates to what you know
MHS is the physical-world counterpart to MCP: where MCP standardizes how an agent discovers and calls software tools, MHS standardizes how an agent discovers and calls hardware. It also extends "computer use" (GUI agents) one layer further, from operating a screen to operating a machine, and its built-in operational constraints are a concrete instance of guardrails applied at the hardware-interface level rather than the prompt level.