Anthropic's MHS lets AI agents control lab equipment and factory robots

Anthropic announced on August 28 the Model Hardware Standard (MHS), an experimental framework that enables AI agents to connect directly with physical hardware without complex middleware. The standard — described as a physical-world counterpart to Anthropic's Model Context Protocol (MCP) — reduces the time to integrate a new lab instrument or industrial machine from weeks to hours, the company said.

MHS provides a universal translation layer between AI agents and hardware: standardized drivers expose each device's capabilities through simple "read" and "write" commands, along with safety parameters and operational limits. AI agents access these drivers via MCP, making the standard model-agnostic and compatible with any AI system, not just Anthropic's Claude.

What it enables

The initial target environments are scientific research laboratories and advanced manufacturing. In a research lab, an AI agent using MHS could operate a liquid handler, a microscope, a robotic arm, and a quantum computing laser system — coordinating autonomous 24-hour experiments, adjusting parameters in real time, and recovering from certain hardware errors without human intervention. In manufacturing, the same approach could coordinate multi-machine workflows across an entire production line.

Partners supporting the MHS launch include AWS, Genentech, Universal Robots, Doosan Robotics, QIAGEN, Hugging Face, and Raspberry Pi — spanning cloud infrastructure, biotech, industrial robotics, and edge computing. These partners are working with Anthropic to develop safety evaluations and best practices for AI systems operating physical equipment.

MHS vs. MCP

Anthropic's MCP, released in late 2024, became a widely adopted standard for connecting AI agents to software tools, APIs, and data sources. MHS extends that same philosophy into the physical world. Where MCP abstracts over software, MHS abstracts over hardware — providing a consistent interface regardless of whether the device is a lab centrifuge or a six-axis industrial robot.

MHS is currently in a research preview phase, with an open-source release planned for a later date. The announcement signals Anthropic's ambition beyond conversational AI: building the infrastructure for AI agents to operate autonomously in physical environments, a space where competitors including Google DeepMind (with robotics research) and Figure AI are also active.

The standard will face scrutiny on safety — autonomous AI control of physical equipment carries inherently higher stakes than software automation. Anthropic said its partner network is focused on safety evaluations before broader deployment, as first reported on the Anthropic blog.