Model Hardware Standard puts Claude in control of lab rigs

Anthropic has opened a research preview of the Model Hardware Standard, a shared specification for AI agents to operate physical devices, cutting integration work that typically takes weeks or months down to hours or minutes. The first group of participants covers scientific research labs and advanced manufacturers, according to Anthropic.

At a glance

• The specification ships a standardized driver built on read and write primitives, such as get temperature and set temperature, and makes each device discoverable so agents and instruments find each other across networks.
• Natural-language tags let users describe traits code cannot show, such as a robot arm's weight, and the driver compiles them into a reference file listing measurements, adjustable parameters and enforced safety limits.
• Early partners report gains: Carnegie Mellon ran dose-response experiments about three times faster, and a QuEra agent recovered a quantum machine's laser lock 99.3% of the time without human intervention.

Lab and factory automation has long been a per-site engineering project, with each vendor's interface stitched together by specialists and the result fragile by construction. MHS reads as an attempt to do for instruments what MCP did for software tools: one discovery format, one command vocabulary, and safety limits declared at the device rather than buried in a script. The commercial signal is arguably the stronger one, since instrument vendors are building support before the standard is public.

The MHS driver exposes read and write primitives plus natural-language tags

Each instrument normally ships with its own programming interface, and linking several of them requires bespoke translator software. The MHS driver replaces that with primitives such as read (get temperature) and write (set temperature) that any programmable device can act on, plus a standard discovery format.

Details code cannot expose, such as the weight of a robot arm, have lived in paper manuals or as tacit knowledge. Driver tags let users write them in plain language, either directly or through an agent that interviews them, and the driver compiles a reference file covering measurable values, adjustable parameters and enforced safety limits.

Control runs through three mechanisms: MCP, the command line interface and code files, which together allow orchestration across instruments from a single line of code. The standard is model-agnostic, and for long-running or fast operations the agent chains driver commands into code files so hardware executes without a model reasoning at each step.

A QuEra agent recovers the laser lock 99.3% of the time without human intervention

QuEra, which builds quantum computers using neutral atoms, gave an agent control over parts of the laser system inside its machines. The agent developed a controller that recovers the laser's lock, the ultra-precise frequency the lasers must hold to interact with the atoms, 99.3% of the time without human intervention.

Genentech tested MHS as a proof of concept for automating the BCA protein assay, coordinating a liquid handler, a robotic arm and a plate reader. At Carnegie Mellon, an agent ran serial dilution dose-response experiments about three times faster than before, across three computers with fundamentally incompatible interfaces.

In the University of Washington Baker and Pinglay labs, PhD student Zihao Song built a remote instrument dashboard, an agent-supervised qPCR that halts on the amplification curve, and collision-free plate handoffs between a robotic arm and a liquid handler. Tetsuwan Scientific ran a qPCR workflow profiling pollution in San Pedro Creek.

MHS began at HHMI Janelia, and vendors from AWS to QIAGEN are adding support

MHS started as work between Alek Kemeny of Anthropic's Beneficial Deployments team and Arco Bast, a postdoctoral scientist at HHMI Janelia, whose brain-imaging rig combined lasers, motorized focusers and cameras from different vendors with no common interface. Bast built a shared memory dictionary to let the instruments talk, and the two then integrated AI models into that interface.

At Janelia, Virginie Ruetten of the Ahrens lab used MHS to unify a rig that previously required seven separate vendor programs. Amazon Web Services will support the standard through Strands Robots and is providing a private pre-release version of the package for the duration of the preview; Tecan is adding support for its Fluent liquid handlers.

Automata is adding MHS to its LINQ platform for instrument error handling, MBF Bioscience is building a driver for ScanImage, and QIAGEN has a working proof of concept on QIAsymphony Connect. Doosan Robotics is testing MHS with robotic arms, Universal Robots plans platform support, and Danaher is exploring the standard.

Ahead of the open-source release

Anthropic has not given a date for open-sourcing MHS, and says findings from the preview will be published as deployment guidance. Spatial and physical reasoning remain limits: Genentech researchers had to guide Claude to treat sample foaming as a physical failure rather than a software bug.

Hardware without a programming interface is not covered, and manufacturers of such devices are being enlisted to build drivers in. Hugging Face is adding MHS support to LeRobot, and Raspberry Pi is enabling integration across several products after tests with its Camera MHS Driver. Access runs through a waitlist.