nvidia-isaac-mcp
Enables AI tools like Windsurf and Claude to control NVIDIA Isaac Sim and Isaac Lab through natural language, providing tools for scene inspection, prim management, physics simulation, and robot spawning.
README
NVIDIA Isaac MCP
MCP (Model Context Protocol) server for NVIDIA Isaac Sim & Isaac Lab — enables AI tools like Windsurf, Claude Desktop, and Claude Code to control Isaac Sim through natural language.
Architecture
AI Client (Windsurf/Claude) ←— MCP (stdio) —→ nvidia-isaac-mcp (this package)
│
TCP Socket :9876
│
isaacsim.mcp.server
(Extension inside Isaac Sim)
Two components:
- Isaac Sim Extension (
exts/isaacsim.mcp.server/) — runs inside Isaac Sim, listens on TCP port 9876 - MCP Server (
src/nvidia_isaac_mcp/) — external Python process that AI tools connect to
Prerequisites
- NVIDIA Isaac Sim 5.x
- Python 3.10+
- uv (recommended)
Setup
1. Start Isaac Sim with the extension
cd <your-isaac-sim-install>
bash isaac-sim.sh \
--ext-folder <path-to-nvidia-isaac-mcp>/exts \
--enable isaacsim.mcp.server
The extension will start listening on 127.0.0.1:9876.
2. Run the MCP server
cd <path-to-nvidia-isaac-mcp>
uv run nvidia-isaac-mcp
3. Configure your AI tool
Windsurf / Claude Desktop — add to MCP settings:
{
"mcpServers": {
"isaac-sim": {
"command": "uv",
"args": ["run", "--directory", "<path-to-nvidia-isaac-mcp>", "nvidia-isaac-mcp"]
}
}
}
Available Tools
Core
| Tool | Description |
|---|---|
ping |
Check connectivity to Isaac Sim and return server version |
Scene Inspection
| Tool | Description |
|---|---|
get_stage_info |
Overview of the current USD stage (prims, up-axis, etc.) |
get_prim_info |
Detailed info about a specific prim (attributes, transform, children) |
Prim CRUD
| Tool | Description |
|---|---|
create_prim |
Create a USD prim (Cube, Sphere, Cylinder, Cone, Plane, Capsule, Xform) |
modify_prim |
Modify a prim's transform or visibility |
delete_prim |
Remove a prim from the stage |
Code Execution
| Tool | Description |
|---|---|
execute_code |
Run arbitrary Python code inside Isaac Sim's runtime |
Physics & Simulation
| Tool | Description |
|---|---|
create_physics_scene |
Create a PhysicsScene prim to enable simulation |
add_rigid_body |
Add rigid body physics to an existing prim |
add_ground_plane |
Add a physics-enabled ground plane |
simulation_control |
Play, pause, or stop the simulation |
Robots & Assets
| Tool | Description |
|---|---|
spawn_robot |
Spawn a built-in robot (franka, ur10, jetbot, etc.) |
list_available_robots |
List available built-in robot assets |
load_usd_asset |
Load any USD file as a reference into the stage |
Testing
# Direct TCP test (no MCP server needed, just the extension)
python tests/test_e2e.py
# MCP inspector
uv run mcp dev src/nvidia_isaac_mcp/server.py
Environment Variables
| Variable | Default | Description |
|---|---|---|
ISAAC_SIM_HOST |
127.0.0.1 |
Extension socket host |
ISAAC_SIM_PORT |
9876 |
Extension socket port |
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。