Isaac MCP

The Isaac MCP (Model Context Protocol) integration lets AI agents interact directly with NVIDIA Isaac Sim and Isaac Lab environments. This enables LLM-driven robotics — an AI agent can spawn simulations, reset environments, adjust parameters, and observe results through a standardized protocol.

What is Isaac MCP?

MCP (Model Context Protocol) is a standard for connecting AI models to external tools. Our Isaac MCP server exposes Isaac Sim/Lab as a set of tools that any MCP-compatible agent can call:

• Spawn environments — create simulation scenes with the Dropbear robot
• Run training — launch RL training loops with configurable hyperparameters
• Observe — stream joint states, IMU, camera feeds from the sim
• Deploy policies — load trained neural network policies into the sim
• Reset & configure — modify terrain, physics, robot parameters on the fly

Architecture

The Isaac MCP server runs alongside Isaac Sim and exposes tools over the MCP protocol. Any MCP client (Claude, Hermes, custom agents) can connect and drive the simulation.

Getting Started

The Isaac MCP server is part of the dropbear_isaac repository.

# Clone the Isaac integration
git clone https://github.com/Hyperspawn/dropbear_isaac.git
cd dropbear_isaac

# Install dependencies (requires Isaac Sim 4.0+)
pip install -e ".[mcp]"

# Start the MCP server
python -m dropbear_isaac.mcp_server --port 8765

Use Cases