Reason over context
The large model interprets the task, device state, and constraints.
MHS PROJECTS · HARDWARE · MCP SERVERS
MHS is a proposed standard for how an AI Model or Agent discovers, understands, and controls Hardware. Find MHS projects, compatible open hardware, and MCP servers here.
MHS is currently a limited research preview. The core specification is not yet public open source.
SourceTHE PHYSICAL AI STACK
Four terms explain the stack. The model reasons, the agent acts, MCP exposes software tools, and MHS describes the physical hardware layer.
The large model interprets the task, device state, and constraints.
The agent turns model output into tool calls and checks results.
MCP servers expose tools, resources, and services to the agent.
Browse MCP Servers →MHS aims to describe devices, state, operations, and safety limits.
Browse Hardware →CURATED DIRECTORY
Start with projects connected to MHS and hardware that already supports AI model training or agent control.
Strands Labs
Control robots and physical hardware with natural language through Strands Agents.
Hugging Face
An open-source library for end-to-end robot learning and real-world robotics.
The Robot Studio
Open, build-it-yourself robot arms designed to work with the LeRobot AI robotics library.
MODEL HARDWARE STANDARD
MHS aims to replace one-off hardware integrations with a common description of devices, state, operations, and safety limits.
What is MHS?The Agent can discover a device and the operations it exposes.
The Model receives structured state and operating context before acting.
The Hardware driver enforces limits outside the Model prompt.
MHS + MCP
MCP connects a Model or Agent to software tools and data. MHS focuses on physical devices, their state, available operations, and safety boundaries. A hardware driver can be exposed through MCP, a CLI, or a code API.