Anthropic previews new standard to let AI agents control lab machines

The effort addresses a practical challenge: scientific instruments each have their own programming interfaces, often incompatible, so researchers must write custom code to coordinate devices. MHS replaces these disparate APIs with a single, unified interface and a common set of configuration commands, allowing machines to exchange data over a network without bespoke translator programs. Anthropic said the approach could cut experiment setup from weeks or months to hours or minutes.

According to SiliconANGLE, Anthropic developed MHS with HHMI, which has built an AI agent that automatically manages microscopes. Startup QuEra Computing is using MHS and Claude to coordinate lasers for programming quantum computers. The standard also lets agents automatically fix certain experiment errors and generate control scripts; in one internal test, Claude produced a script that adjusts laser positions at lower cost than inference-based configuration.

Researchers can further benefit from files describing instrument properties, such as a robotic arm's weight or a sensor's wavelengths, helping agents use each machine correctly. Labs can also define safety guardrails, such as limits on laser strength to avoid sample damage.

Anthropic plans to release MHS under an open-source license and says it could eventually apply to manufacturing equipment as well as science. Amazon Web Services, Hugging Face and several industrial robot suppliers are already working to integrate the standard, according to SiliconANGLE. Ars Technica noted that the project was inspired by a neuroscientist at HHMI Janelia Research Campus who had designed a custom interface to coordinate lasers, microscopes and cameras in an experiment.