iflow-mcp_houdini-mcp 0.1.0

Connect Houdini to Claude via Model Context Protocol

HoudiniMCP – Connect Houdini to Claude via Model Context Protocol

HoudiniMCP allows you to control SideFX Houdini from Claude using the Model Context Protocol (MCP). It consists of:

1. A Houdini plugin (Python package) that listens on a local port (default localhost:9876) and handles commands (creating and modifying nodes, executing code, etc.).
2. An MCP bridge script you run via uv (or system Python) that communicates via stdin/stdout with Claude and TCP with Houdini.

Below are the complete instructions for setting up Houdini, uv, and Claude Desktop.

---

Table of Contents

1. Requirements
2. Houdini MCP Plugin Installation
   1. Folder Layout
   2. Shelf Tool (Optional)
   3. Packages Integration (Optional)
3. Installing the mcp Python Package
   1. Using uv on Windows
   2. Using pip Directly
4. Bridging Script and Claude for Desktop
   1. The Bridging Script
   2. Telling Claude Desktop to Use Your Script
5. Testing & Usage
6. Troubleshooting

---

Requirements

• SideFX Houdini
• uv
• Claude Desktop (latest version)

---

1. Houdini MCP Plugin Installation

1.1 Folder Layout

Create a folder in your Houdini scripts directory: C:/Users/YourUserName/Documents/houdini19.5/scripts/python/houdinimcp/

Inside houdinimcp/, place:

• __init__.py – handles plugin initialization (start/stop server)
• server.py – defines the HoudiniMCPServer (listening on port 9876)
• houdini_mcp_server.py – optional bridging script (some prefer a separate location)
• pyproject.toml

(If you prefer, houdini_mcp_server.py can live elsewhere. As long as you know its path for running with uv.)

1.2 Shelf Tool

create a Shelf Tool to toggle the server in Houdini:

1. Right-click a shelf → "New Shelf..."

Name it "MCP" or something similar

2. Right-click again → "New Tool..." Name: "Toggle MCP Server" Label: "MCP"

3. Under Script, insert something like:

   import hou
   import houdinimcp

   if hasattr(hou.session, "houdinimcp_server") and hou.session.houdinimcp_server:
       houdinimcp.stop_server()
       hou.ui.displayMessage("Houdini MCP Server stopped")
   else:
       houdinimcp.start_server()
       hou.ui.displayMessage("Houdini MCP Server started on localhost:9876")

1.3 Packages Integration

If you want Houdini to auto-load your plugin at startup, create a package file named houdinimcp.json in the Houdini packages folder (e.g. C:/Users/YourUserName/Documents/houdini19.5/packages/):

{
  "path": "$HOME/houdini19.5/scripts/python/houdinimcp",
  "load_package_once": true,
  "version": "0.1",
  "env": [
    {
      "PYTHONPATH": "$PYTHONPATH;$HOME/houdini19.5/scripts/python"
    }
  ]
}

2 Using uv on Windows

  # 1) Install uv 
  powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

  # 2) add uv to your PATH (depends on the user instructions) from cmd
  set Path=C:\Users\<YourUserName>\.local\bin;%Path%

  # 3) In a uv project or the plugin directory
  cd C:/Users/<YourUserName>/Documents/houdini19.5/scripts/python/houdinimcp/
  uv add "mcp[cli]"

  # 4) Verify
  uv run python -c "import mcp.server.fastmcp; print('MCP is installed!')"

3 Telling Claude for Desktop to Use Your Script

Go to File > Settings > Developer > Edit Config > Open or create: claude_desktop_config.json

Add an entry:

{
  "mcpServers": {
    "houdini": {
      "command": "uv",
      "args": [
        "run",
        "python",
        "C:/Users/<YourUserName>/Documents/houdini19.5/scripts/python/houdinimcp/houdini_mcp_server.py"
      ]
    }
  }
}

if uv run was successful and claude failed to load mcp, make sure claude is using the same python version, use:

  python -c "import sys; print(sys.executable)"

to find python, and replace "python" with the path you got.

4 Use Cursor

Go to Settings > MCP > add new MCP server add the same entry in claude_desktop_config.json you might need to stop claude and restart houdini and the server

5 OPUS integration

OPUS provide a large set of furniture and environmental procedural assets. you will need a Rapid API key to log in. Create an account at: RapidAPI Subscribe to OPUS API at: OPUS API Subscribe Get your Rapid API key at OPUS API add the key to urls.env

4 Acknowledgement

Houdini-MCP was built following blender-mcp. We thank them for the contribution.