Anthropic has introduced an interface designed to make it easier for AI agents to operate machines in the real world, marking the AI lab's continued foray into physical AI.

The generative AI vendor unveiled a preview of the Model Hardware Standard (MHS) last week, with initial deployments set for scientific research and advanced manufacturing tools, including microscopes, robotic arms and other lab equipment.

Anthropic developed the standard in collaboration with the private biomedical research institute HHMI.

Beyond operating instruments, Anthropic said MHS enables agents to perform complex tasks, including drug discovery and laser calibration on a quantum computer.

Automating these tasks typically requires custom coding, but with MHS, Anthropic says it can standardize the process and reduce integration time to hours or even minutes. The standard is also model agnostic, meaning users are not locked into Anthropic's Claude models.

MHS can also provide agents with information about hardware that may not be apparent from code alone, such as a robotic arm's weight or safety limits. Users can supply this information in natural language, after which MHS creates a reference file that tells the agent what a device can measure, what can be adjusted and how it can be operated safely. That kind of machine-readable safety context addresses a long-standing hurdle in lab and industrial automation, where instruments have traditionally exposed only low-level control interfaces rather than structured descriptions of their capabilities and constraints.

"By incorporating AI into these tools, MHS helps researchers and engineers more readily orchestrate autonomous, round-the-clock experiments and workflows," Anthropic said in a blog post. "Agents [are] able to reason through each step in an experiment, update parameters in real time, and, in some cases, recover from hardware errors without intervention."

Anthropic plans to make MHS available under an open source license, meaning it will be available to any device manufacturer across any industry. In addition to the current scientific applications, MHS is expected to be rolled out in manufacturing, with AWS and Hugging Face already working to integrate the standard into their workflows.

This is not the first AI standard Anthropic has released: the vendor open sourced the Model Context Protocol (MCP) in 2024. Now widely used, MCP is designed to connect AI agents and data sources. MCP's rapid adoption — it has since been embraced by major AI and enterprise software vendors — helped establish a playbook that Anthropic appears to be extending here: release an open interoperability layer, let the ecosystem build on it, and broaden the reach of agentic AI beyond text and code.

MHS signals a further move into the physical world for Anthropic, as AI companies increasingly look to hardware amid rising investment in AI-native devices and manufacturing tools. Early reactions to the preview and the involvement of partners like AWS and Hugging Face suggest how adoption may unfold, but the standard's success will depend on whether device manufacturers choose to implement it.