Anthropic Launches Model Hardware Standard (MHS), Enabling Claude to Directly Control Physical Hardware
量子位 · wechat · 2026-08-28
Anthropic has released the Model Hardware Standard (MHS), a driver and device description standard for hardware, acting as a physical-world counterpart to the Model Context Protocol (MHS). MHS translates diverse hardware devices—such as robotic arms, microscopes, and lab instruments—into unified interfaces that Agents can discover, read, and invoke, enabling Claude to interact with the physical world without specific training.
Key Capabilities and Demos:
- Direct Robot Control: Claude controlled the SO-ARM101 robotic arm from the Hugging Face LeRobot ecosystem via MHS, automatically measuring the workspace, calibrating, and executing actions without teleoperation or human demonstration.
- Automated Experiments: In a demo tracking cells under a microscope, Claude wrote a tracking program and autonomously controlled the equipment, freeing scientists from tedious manual operation.
- Universal Connectivity: It unifies devices from different vendors (e.g., lasers, cameras) that previously had isolated data formats.
Technical Principle:
MHS provides a standardized driver and a machine-readable “manual” for each device, detailing its current state, executable actions, and safety boundaries. Claude discovers devices via MCP or code and makes decisions based on real-time feedback.
Currently in a research preview available to select institutions, MHS is planned to be open-sourced after safety assessments. This move extends Anthropic’s vision from connecting digital tools (MCP) to connecting physical entities, exploring a path towards “distributed embodiment.”
More from Embodied
- How Microsoft optimized Windows for Nvidia's Arm-based RTX Spark laptops — The Verge AI · 2026-10-15
- John Deere's second-gen autonomous tillage: 16 cross-checking cameras beat sensor drift — lukas_m_ziegler · 2026-10-15
- Europe's Application-Layer Bet: Can the Rotterdam Humanoid Center Deliver? — ingliguori · 2026-10-15
- VC thesis 'Data at the Edge': cheap sensors plus AI are rebuilding the internet's data flywheel offline — rebeccakaden · 2026-10-15
- PingpongDuo shows humanoid robots playing doubles table tennis — arankomatsuzaki · 2026-10-15
- Robotics data gap: just 300K hours of manipulation data globally vs 100M+ needed, $3B spend projected — 0xSammy · 2026-10-15