Anthropic opens its Model Hardware Standard

Anthropic has launched a research preview of the Model Hardware Standard, or MHS, aimed at making physical lab and manufacturing equipment easier for AI agents to control safely. The effort targets a familiar bottleneck in robotics and scientific automation: every device tends to require custom integration work before an agent can use it. By standardizing how hardware is described and controlled, Anthropic is positioning MHS as infrastructure for agentic workflows in labs, factories, and other programmable environments. The preview is initially going to research labs and advanced manufacturers before a broader open-source release.

Key Features or Updates

MHS provides a shared driver model and device description layer so agents can discover, reason about, and operate different instruments through a common interface. Anthropic says the standard is model-agnostic and can work with standard protocols such as MCP, while also carrying natural-language safety and capability metadata for each device.

Impact on Developers

For developers building agent systems around robotics, biotech, or industrial workflows, the headline is reduced integration drag. Instead of stitching together one-off adapters for each microscope, robot arm, or liquid handler, teams could target a common abstraction and spend more time on orchestration, verification, and recovery logic.

How to use it

The preview is geared toward research and industrial partners that already manage programmable devices and want to test agent-driven automation. Teams can evaluate whether their hardware can expose the required programmable interfaces, then use MHS as the standard layer between agent harnesses and physical equipment.