Anthropic Releases Research Preview of MHS Hardware Standard for AI Agents

Anthropic has released a research preview of the Model Hardware Standard (MHS) targeting scientific laboratories and advanced manufacturers. MHS is a shared specification that allows AI agents to safely operate various experimental and manufacturing devices, such as microscopes, liquid handlers, and robotic arms, in parallel.

Key Features of MHS

Previously, hardware integration took weeks to months, but MHS reduces this process to hours or minutes. It translates communication between the OS and hardware through standardized drivers and controls devices using basic commands (primitives) such as 'read' and 'write'.

Additionally, MHS drivers make devices identifiable in a standard format, allowing devices and agents on a network to discover each other without separate translation programs. It automatically generates reference files containing device characteristics, measurable items, and safety limits based on tags entered by users in natural language, helping agents understand the devices.

Control Methods and Use Cases

MHS controls devices via MCP (Model Context Protocol), command-line interfaces, and code files (APIs). This allows agents to orchestrate multiple devices with a single line of code, execute experimental steps sequentially, and adjust parameters in real time. For long-running tasks or high-speed operations, driver commands can be chained via code files, allowing devices to perform tasks autonomously without real-time inference from the agent.

MHS is compatible with any device that has a programmable interface and can be accessed via a standard protocol regardless of the model (any model-agnostic). Anthropic is collaborating with partners in science, robotics, electronics, and manufacturing to develop safety evaluations and best practices ahead of its open-source release.