Anthropic has opened a research preview of the Model Hardware Standard (MHS), a shared specification that lets Claude and other AI agents safely operate laboratory and manufacturing hardware, the company said in a blog post on August 27, 2026.
Fixing a Fragmented Integration Problem
Most lab and factory equipment — liquid handlers, robotic arms, plate readers, microscopes, qPCR machines — has no common interface, so connecting an AI agent to new hardware traditionally requires specialists to build custom integrations over weeks or months, according to Anthropic. MHS standardizes that work around shared “read” and “write” primitives and natural-language tags describing how a device should be operated safely. Agents can reach devices through the Model Context Protocol, direct command-line interfaces, or code files for chaining multi-step workflows. Anthropic says the standard cuts integration time to “hours or minutes.”
Early Results From Six Labs
Anthropic tested MHS with six organizations ahead of the preview. Genentech used it to automate protein-assay handoffs between a liquid handler, a robotic arm and a plate reader. Carnegie Mellon University ran serial-dilution dose-response experiments three times faster, while the University of Washington’s Baker and Pinglay labs used MHS to monitor qPCR runs and coordinate robotic plate transfers. HHMI Janelia consolidated seven separate vendor programs into one microscopy control layer, and QuEra Computing built laser-stabilization controllers that reached a 99.3% autonomous-recovery rate. Tetsuwan Scientific used MHS to run qPCR-based pollution-monitoring workflows in California.
Hardware makers including Amazon Web Services, Danaher, Doosan Robotics, Tecan, QIAGEN and Universal Robots are adding MHS support, Anthropic said. The push extends a broader effort to put Claude to work on physical and scientific tasks, following the company’s report that Claude designed working protein binders in lab tests, and it lands alongside a wider industry push to standardize how AI agents talk to tools and to each other.
Still a Limited Preview
Access remains restricted to research labs and manufacturers that apply through Anthropic’s waitlist; a full open-source release is planned once the research preview concludes and safety practices are finalized. The cautious rollout mirrors how Anthropic has staged other Claude capabilities as it pushes further into hardware and lab automation.