Quick Facts
- Anthropic’s Model Hardware Standard reduces hardware integration time from months to hours or minutes, according to the company.
- Early partners include Genentech, QuEra, Carnegie Mellon University, AWS, Danaher, Universal Robots, and Hugging Face.
- The standard is model-agnostic and designed to extend Anthropic’s existing Model Context Protocol into the physical world.
Anthropic on Aug. 27 previewed the Model Hardware Standard, a new interface specification that lets AI agents operate physical machinery including robotic arms, microscopes, liquid handlers, and laser systems. The standard targets laboratories and manufacturing facilities where hardware from different vendors typically requires months of custom integration work before it can be automated.
The project grew out of a collaboration between Anthropic’s Beneficial Deployments team and a postdoctoral researcher at HHMI Janelia Research Campus. That researcher had built a brain-imaging rig combining lasers, motorized focusers, and cameras from multiple vendors with no shared interface. His workaround, a shared memory dictionary allowing instruments to communicate at memory speed, became the foundation for MHS.
MHS extends Anthropic’s existing Model Context Protocol, which governs how AI models communicate with software tools and data sources. MHS applies the same logic to hardware, giving devices a common driver interface, natural-language device tags, and auto-generated reference files that detail operational parameters and safety limits.
Early results from pilot deployments are striking. QuEra, a quantum computing company, used a Claude-built laser relock system that achieved a 99.3% success rate across 700 trials, up from 58% with a custom script. HHMI Janelia unified seven vendor programs into a single interface and cut new hardware integration time from multiple days to minutes. The University of Washington connected six lab instruments in under a week, replacing months of traditional automation work, and eliminated the need for manual plate swapping every 90 minutes during qPCR runs.
Genentech engineers deployed MHS across an automated protein assay workflow where Claude autonomously optimized liquid transfer rates for varying viscosities. In one case, Claude repeatedly attempted to clear fluid-handling bubbles by retrying the same software command, requiring human intervention. The incident illustrates a key limitation: AI models learn about the physical world primarily through text and images, leaving gaps in spatial reasoning that still require expert oversight.
Elizabeth Kelly, Head of Beneficial Deployments at Anthropic, said: “We built this for science to sort of show the promise of AI, but there’s also huge benefits here for enterprise and for industry.”
Jonah Cool, Head of Partnerships and Deployment of Science at Anthropic, pointed to the fragmentation problem MHS is designed to solve. Scientific equipment, he said, “suffers from proprietary solutions that are very brittle and often don’t meet the need of scientists.” The goal, he added, is to “avoid vendor lock-in for scientists.”
Safety constraints are built into the protocol itself rather than delegated entirely to the AI model. Speed limits and angle restrictions for robotic systems are embedded at the interface level, so an agent cannot instruct a device to move outside pre-set safe parameters. Anthropic says model-level guardrails are also designed to block misuse, including attempts to synthesize dangerous compounds through automated lab equipment. The company is developing a physical safety roadmap and plans to release research findings alongside any future open-source release.
MHS is model-agnostic. Companies adopting it are not locked into Claude. Any AI system compatible with the specification can, in theory, operate the same hardware.
On the vendor side, AWS plans to support MHS through its Strands Robots library. Automata expects to add MHS to its LINQ lab automation platform. Danaher, Doosan Robotics, MBF Bioscience, Qiagen, Tecan, and Universal Robots have also signaled planned support. Hugging Face is planning integration with its LeRobot project, and Raspberry Pi is testing driver compatibility.
Anthropic has not announced a timeline for open-sourcing the standard. “There’s more to learn before we open source MHS,” the company said in its announcement.
Read more: Anthropic previews MHS standard for AI agents that operate machines