Anthropic has previewed the Model Hardware Standard (MHS), a unified interface that lets Claude and other AI agents discover, describe and control physical machines — from robot arms and microscopes to liquid handlers and quantum-computer lasers — through a single, standardized API.
Fixing the ‘every instrument has its own SDK’ problem
Scientific and industrial equipment usually ships with device-specific SDKs that mix Python, Java and vendor-only protocols, forcing researchers to write brittle glue code. MHS replaces those disparate APIs with a shared discovery layer and a common set of configuration commands. It also stores each device's physical characteristics — weight, safety limits, adjustable parameters — that previously lived in paper manuals or the heads of specialists.
Built with HHMI, tested with Genentech and QuEra
Anthropic developed MHS with the medical research institute HHMI, which used it to build an agent that automatically manages microscopes at its Janelia Research Campus — compressing an imaging experiment from weeks to a single day. Genentech ran a drug-discovery workflow with real-time error handling, and quantum startup QuEra Computing is now using Claude and MHS to coordinate the lasers that program its neutral-atom quantum computer.
Safety guardrails baked into the driver
Because MHS runs as a standardized driver, labs can encode limits on the strength of laser beams or force from a robot arm directly in the device description. Anthropic says agents can now automatically write instrument control scripts — in one internal test, Claude generated a script that adjusted laser positions on the fly, cheaper than an inference-driven configuration loop.
Open source, with hyperscaler and robotics backers
The company plans to release MHS under an open-source license. Amazon Web Services, Hugging Face and several industrial robot suppliers are already integrating it. The move sits alongside a broader push into physical AI, including Hugging Face's $399 Microduck launch, Anthropic's $45B Nscale compute deal and last week's Pentagon court win.
Reporting based on coverage from SiliconANGLE, CNBC, The Register, Fortune, Quartz and Anthropic's research preview announcement.