Isaac Simulation MCP Extension and Server – adds natural-language, MCP-based control of NVIDIA Isaac Sim via a standalone Python server and an Omniverse extension.
https://github.com/omni-mcp/isaac-sim-mcpStop wrestling with Isaac Sim's complex APIs just to place a robot or adjust lighting. This MCP server transforms your AI assistant into a direct control interface for NVIDIA's powerful robotics simulator.
You know Isaac Sim is incredibly capable for robotics simulation, but getting anything done requires diving deep into USD schemas, Omniverse APIs, and verbose Python scripts. Just creating a simple robot scene involves dozens of lines of boilerplate code.
This MCP server changes that completely. Instead of writing simulation scripts, you describe what you want in plain English through your AI assistant, and it happens instantly in Isaac Sim.
Instant Robot Deployment: "Create 3x3 Franka robots across locations [3,0,0] and [6,3,0]" spawns a full robot array without writing a single line of positioning code.
Natural Scene Building: "Add more light to the stage and create obstacles for navigation testing" handles environment setup that would normally require multiple API calls and parameter tuning.
Multi-Robot Orchestration: Control different robot types (Franka, G1, Go1, Carter, Jetbot) simultaneously with conversational commands instead of managing separate control loops.
Real-time Physics Tweaking: Adjust gravity, lighting, and physics parameters through chat instead of diving into simulation settings panels.
Rapid Prototyping: Test robot behaviors and scene configurations in seconds rather than spending time coding simulation setups.
Research Iteration: Focus on algorithm development instead of simulation infrastructure - describe your test scenario and iterate immediately.
Demo Preparation: Build impressive multi-robot demonstrations through conversation, perfect for stakeholder presentations or research showcases.
Educational Use: Students can experiment with complex robotics simulations without mastering Isaac Sim's steep learning curve first.
Instead of this Isaac Sim script:
from omni.isaac.core import World
from omni.isaac.franka import Franka
import numpy as np
world = World()
# 20+ lines of robot positioning, physics setup, lighting configuration...
You simply tell your AI assistant:
Create a physics scene with 3 Franka robots positioned at [0,0,0], [2,0,0], and [4,0,0].
Add proper lighting and make the middle robot move to [2,2,0].
The MCP server handles all the underlying Isaac Sim complexity while you focus on your actual robotics work.
Integrates directly with Cursor AI (or any MCP-compatible editor) while Isaac Sim runs normally in the background. The extension architecture means zero disruption to your current Isaac Sim workflow - it just adds natural language control on top.
The dual-component design (Isaac Sim extension + standalone MCP server) ensures reliable communication between your AI assistant and the simulation environment, even during heavy computational loads.
Bonus: Includes Beaver3D integration for generating 3D objects from images and USD asset search functionality, turning your AI assistant into a complete simulation content creation tool.
If you're doing any serious robotics simulation work with Isaac Sim, this MCP server eliminates the friction between your ideas and implementation. Your AI assistant becomes your simulation engineer, handling the tedious setup while you focus on the robotics problems that actually matter.