Houdini MCP Server
Enables AI-assisted 3D content creation by allowing Claude Code to control SideFX Houdini directly, including creating nodes, setting parameters, executing HScript commands, and building complex procedural networks.
README
Houdini MCP Server
Control SideFX Houdini directly from Claude Code using the Model Context Protocol (MCP).
This project enables AI-assisted 3D content creation by allowing Claude to create nodes, set parameters, execute HScript commands, and build complex procedural networks in Houdini.
Features
🚀 Unlimited Power - Do ANYTHING in Houdini!
This tool gives Claude full access to Houdini's capabilities:
- ALL Contexts: SOPs, DOPs, COPs, CHOPs, VOPs, ROPs, LOPs - Everything!
- Any Node Type: Geometry, simulations, compositing, animation, rendering
- Full HScript Access: Execute any Houdini command with
execute_hscript - VEX Code: Write custom procedural logic inline
- Simulations: Particles, fluids, pyro, cloth, RBD, crowds
- Complex Networks: Build intricate procedural systems
- Real-time Control: Execute commands while Houdini is running
- Natural Language: Just describe what you want in plain English
Core MCP Tools
- Node Management: Create, delete, and connect nodes programmatically
- Parameter Control: Set and animate node parameters with expressions
- Scene Inspection: Query scene structure and node information
- HScript Execution: Run ANY HScript command directly in Houdini
- Network Building: Connect complex node chains automatically
Architecture
The system consists of two components:
- MCP Server (
houdini_mcp_server.py): Runs as a Claude Code MCP server, exposes tools to Claude - Houdini Plugin (
houdini_plugin.py): Runs inside Houdini, listens for commands via TCP socket
┌─────────────┐ ┌──────────────────┐ ┌─────────────┐
│ Claude Code │ ◄─MCP──►│ MCP Server │◄──TCP──►│ Houdini │
│ │ │ (stdio) │ :9876 │ Plugin │
└─────────────┘ └──────────────────┘ └─────────────┘
Requirements
- Houdini 19.5 or later (tested with Houdini 21)
- Python 3.10+ (included with Houdini)
- Claude Code CLI
- Operating System: macOS, Linux, or Windows
Installation
1. Clone the Repository
git clone https://github.com/atayilgun/Houdini-claudecode-mcp.git
cd Houdini-claudecode-mcp
2. Run Setup Script
./setup.sh
The setup script will:
- Detect your Houdini installation
- Install Python dependencies
- Configure Claude Code with the MCP server
3. Start the Houdini Plugin
- Open Houdini
- Go to Windows → Python Shell
- Run these commands in the Python Shell:
import sys
sys.path.append('/path/to/houdini-mcp') # Replace with your actual path
from houdini_plugin import HoudiniMCPServer
server = HoudiniMCPServer()
server.start()
You should see:
✅ Houdini MCP Server listening on localhost:9876
Ready to receive commands from Claude Code!
Usage
Once the plugin is running, you can control Houdini from Claude Code using natural language:
Basic Examples
Create objects:
Create a cube in Houdini
Multiple objects:
Create a sphere at position (5, 0, 0) and a cube at the origin
Animation:
Make the cube move from X=0 to X=10 over 48 frames
Complex procedural setups:
Create a particle system with 5000 points scattered on a sphere,
add colorful gradients, and make them flow like wind
Available MCP Tools
Claude has access to these tools:
create_node(node_type, node_name, parent)- Create a new nodedelete_node(node_path)- Delete a nodeconnect_nodes(source_path, dest_path)- Connect nodesset_parameter(node_path, param_name, param_value)- Set parameter valuesget_scene_info()- Get information about the sceneexecute_hscript(code)- Execute HScript commands
Examples
See the examples/ directory for Python scripts that demonstrate various capabilities:
create_animated_cube.py- Basic animation exampleparticle_system.py- Advanced particle effectsprocedural_network.py- Building complex node networks
Manual Setup (Alternative)
If the setup script doesn't work for your system:
1. Install Dependencies
# Using Houdini's Python
/path/to/houdini/python3 -m pip install fastmcp httpx --user
2. Configure Claude Code
claude mcp add houdini --transport stdio -- /path/to/houdini/python3 /path/to/houdini_mcp_server.py
3. Verify Installation
claude mcp list
claude mcp get houdini
Troubleshooting
Port Already in Use
If you see Address already in use error:
- Make sure no other instance of the plugin is running
- Change the port in both
houdini_mcp_server.pyandhoudini_plugin.py
Cannot Connect to Houdini
- Verify Houdini is running
- Verify the plugin is loaded in Houdini's Python Shell
- Check firewall settings
MCP Server Not Found
# Check if server is registered
claude mcp list
# If not listed, run setup again
./setup.sh
Python Import Errors
Make sure you're using Houdini's Python, not your system Python:
/Applications/Houdini/Current/Frameworks/Python.framework/Versions/Current/bin/python3 --version
Development
Project Structure
houdini-mcp/
├── README.md # This file
├── requirements.txt # Python dependencies
├── setup.sh # Installation script
├── houdini_mcp_server.py # MCP server (runs with Claude)
├── houdini_plugin.py # Plugin (runs in Houdini)
└── examples/ # Example scripts
├── create_animated_cube.py
├── particle_system.py
└── procedural_network.py
Running Tests
Test the connection:
python3 examples/test_connection.py
Adding New Tools
To add a new MCP tool:
- Add the tool decorator and function in
houdini_mcp_server.py:
@mcp.tool()
def your_new_tool(param1: str, param2: int) -> str:
"""Tool description"""
result = send_command({
"type": "your_command",
"params": {"param1": param1, "param2": param2}
})
return f"Result: {result}"
- Add the command handler in
houdini_plugin.py:
def _execute_command(self, command):
cmd_type = command.get("type")
if cmd_type == "your_command":
return self._your_command_handler(command.get("params", {}))
Security Considerations
- The plugin accepts commands from localhost only by default
- HScript execution should be used carefully as it runs code directly in Houdini
- Always review generated code before executing
- Consider network security if opening to non-localhost connections
Contributing
Contributions are welcome! Please:
- Fork the repository
- Create a feature branch
- Make your changes
- Submit a pull request
License
MIT License - See LICENSE file for details
Acknowledgments
- Built with FastMCP
- Powered by Anthropic's Model Context Protocol
- Made for SideFX Houdini
Support
- Issues: GitHub Issues
- Discussions: GitHub Discussions
What Can It Do?
Short answer: EVERYTHING!
The tool already has full access to Houdini through execute_hscript and can:
✅ Create any node in any context (SOPs, DOPs, COPs, CHOPs, VOPs, etc.) ✅ Build simulations (Pyro, FLIP fluids, RBD, Vellum, Wire, etc.) ✅ Composite images (COP networks, color correction, effects) ✅ Animate anything (Keyframes, CHOPs, expressions, motion paths) ✅ Write VEX code (Custom attributes, procedural geometry, deformers) ✅ Manage renders (ROPs, render settings, batch rendering) ✅ Query scene data (Node info, parameter values, geometry attributes) ✅ Execute Python in Houdini (Via HScript's python command)
Example capabilities:
- "Create a pyro explosion simulation"
- "Set up a FLIP water tank with collision objects"
- "Build a procedural building generator with multiple levels"
- "Create a particle system with custom VEX forces"
- "Set up a COP network to add glow and color grade"
Future Enhancements
These are convenience features, not limitations:
- [ ] Direct viewport rendering API (currently use HScript)
- [ ] Binary geometry data transfer (currently use file I/O)
- [ ] Node preset library management
- [ ] Multi-session support for team workflows
- [ ] Web dashboard for monitoring
Happy procedural modeling with AI! 🎨✨
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。