Anthropic Unveils Model Hardware Standard (MHS) to Connect AI Agents to Physical Machines and Robotics
Anthropic announces the Model Hardware Standard (MHS), an open framework extending MCP to allow AI agents to safely operate scientific instruments and robotics.
On August 28, 2026, Anthropic released a research preview of the Model Hardware Standard (MHS)—an open, model-agnostic communication framework that enables AI agents to coordinate and safely operate physical laboratory instruments and industrial robotics. Extending the foundation of the Model Context Protocol (MCP), MHS provides a unified interface for AI models to query device capabilities, verify safety constraints, and execute physical actuation commands across multi-instrument scientific and manufacturing pipelines.
Bridging Digital AI Intelligence with Physical Actuation
While AI agents have demonstrated mastery in digital environments through APIs and code execution, connecting models to physical machinery has traditionally required fragile, proprietary custom drivers. Developed in collaboration with HHMI Janelia Research Campus, MHS establishes an open standard where any physical device with a programmable interface can expose its telemetry, operating boundaries, and executable primitives to AI models.
Key Capabilities & Architectural Highlights
- Unified Hardware Communication: Extends MCP to support real-time physical telemetry, streaming sensor data, and closed-loop robotic feedback.
- Simultaneous Multi-Device Coordination: Coordinates complex multi-stage workflows across microscopes, liquid handlers, precision lasers, and robotic arms.
- Built-In Safety & Limit Guardrails: Hardware devices self-report mechanical tolerances, velocity limits, and safety boundaries to prevent equipment damage.
- Model-Agnostic Open Architecture: Compatible across frontier models (Claude, Gemini, GPT, open-source weights) via standardized client-server bindings.
Unlocking Autonomous Science and Micro-Manufacturing
Early pilot deployments in commercial biotech laboratories demonstrated that MHS-enabled agents can autonomously design, execute, and iterate biochemical assays and quantum computer laser calibrations without continuous human supervision.
Standard Specification Overview
| Feature | Details |
|---|---|
| Protocol Name | Model Hardware Standard (MHS) |
| Developer | Anthropic (with HHMI Janelia Research Campus) |
| Core Integration | Model Context Protocol (MCP) Compatible |
| Target Systems | Robotics, Laboratory Automation, Industrial Actuators, Quantum Rigs |
| License | Open Source with Safety Verification Guidelines |