mcp-registry/isaac-sim-mcp

    ==================
      
       /// MCP ///
      /// ISA ///
        
    ==================
        
    [server:online]
    [protocol:ready]

isaac-sim-mcp

by omni-mcp

Isaac Simulation MCP Extension and Server – adds an MCP-compatible natural-language control layer to NVIDIA Isaac Sim.

69
11
Open Source
GitHub

01

get_scene_info

Ping the Isaac Sim Extension Server to confirm the connection is active and obtain basic scene information.

02

create_physics_scene

Create a physics scene with optional ground plane, gravity vector, custom name, and a list of objects to add.

03

create_robot

Instantiate a robot of a specified type at a given position within the scene.

04

omni_kit_command

Execute an Omni Kit command for creating or manipulating primitives and other scene elements.

05

execute_script

Run arbitrary Python code inside Isaac Sim to perform custom operations.

Installation

Prerequisites
1. NVIDIA Isaac-Sim 2024.x (installed through Omniverse launcher).
2. Python 3.10+ (the interpreter shipped with Isaac-Sim or a system interpreter).
3. git and a C/C++ build tool-chain (needed if native python wheels must be built).
Steps
1. Clone the repository inside your Omniverse/isaac-sim extensions folder or any workspace you prefer:
git clone https://github.com/omni-mcp/isaac-sim-mcp.git
2. (Recommended) Create and activate a dedicated conda/venv environment:
python -m venv .venv source .venv/bin/activate # Windows: .venv\Scripts\activate
3. Install the python package in editable/development mode so that Isaac-Sim can find the extension:
cd isaac-sim-mcp pip install -e .[dev]
4. Register the extension in Isaac-Sim (if not placed in the default "extensions" folder):
- Open Isaac-Sim ➜ Window ➜ Extensions ➜ gear icon ➜ Add Local Folder ➜ point to the repo directory.
5. Enable the extension:
In Isaac-Sim Extension browser, search for "MCP" and toggle the checkbox.
6. Launch the MCP server:
python -m mcp_server --host 0.0.0.0 --port 8080
7. Verify by sending a health ping:
curl http://localhost:8080/health
8. (Optional) Create a systemd/service or Docker container for production use.

Documentation

License: MIT License
Updated 7/30/2025