Anthropic has opened a research preview of the Model Hardware Standard (MHS), a shared specification that enables AI agents to safely operate multiple physical devices—such as microscopes, liquid handlers, and robotic arms—in parallel. The standard emerged from a collaboration between Anthropic and HHMI Janelia Research Campus to solve the chronic problem that integrating laboratory and manufacturing hardware typically takes weeks or months because each device has its own programming interface and there is no common way for them to share data with an AI agent. MHS reduces this integration work to hours or minutes by introducing a standardized driver that translates between a computer’s operating system and the hardware using simple primitives like “read” and “write.” The driver makes each device discoverable in a standard format and includes tags that let users write safety-relevant information—for example, the weight of a robot arm—in natural language, which the driver then compiles into a reference file that gives the agent everything it needs to operate the device safely. Agents control the hardware through three mechanisms: the Model Context Protocol (MCP), a command-line interface, and code files (APIs). Together these allow orhestration across multiple devices with a single line of code. The agent can also chain driver commands into deterministic scripts so that devices can carry out long-running operations without requiring the agent to reason at every step.

Early tests with research and industry partners have demonstated time savings and new capabilities. GeneneTech used MHS to automate a standard protein assay that required cooradinating a liquid handler, robotic arm, and plate reader. The University of Washington Baker and Pinglay labs built a remote monitoring dashboard, AI-supervised qPCR, and collision-free plate handoffs between a robotic arm and liquid handler. Carnegie Mellon University ran dose-response experiments about three times faster than before by orhestrating devices on three incompatibile computers. HHMI Janelia unifified a imaging rig that previously required seven different vendor programs. QuEra Computing used MHS to give an AI agent control over its quantum computer’s laser system; the agent developed a controller that recovers the laser’s ultra-precise frequency lock 99.3% of the time without human intervention. Tetsuwan Scientific integrated MHS with its automated biology lab platform to run qPCR workflows for citizen-science pollution profiling.

Several hardware venders and software companies are building MHS support into their platforms. These include AWS (through Strands Robots), Atuomata (LINQ), Danaher, Doosan Robotics, MBF Bioscience (ScanImage), QIAGEN (QIAsymphony Connect), Tecan (Fluent), and Universal Robots. Hugging Face is adding MHS support in LeRobot, and Raspberry Pi is enabling integration across a number of their products.

The standard is not yet open-source, and Anthropic has noted important limitations. Claude’s spatial and physical reasoning—learned from text and images—still requires expert oversight, as GeneneTech researchers had to guide it to recognize that foaming errors were physical failures, not software bugs. MHS also only works with devices that have a programmable interface, so the team is working with manufacturers of non-programable equipment to build in MHS drivers. Anthropic will use the research preview to develop additional safety evaluations and a physical safety roadmap before making the standard open source.