A Unity MCP (Model Context Protocol) package that enables seamless communication between Unity and local Large Language Models (LLMs) via Ollama. This package extends justinpbarnett/unity-mcp to work with local LLMs, allowing developers to automate workflows, manipulate assets, and control the Unity Editor programmatically without relying on cloud-based LLMs.

The Unity MCP with Ollama Integration provides a bidirectional communication channel between:

1. Unity (via C#)
2. A Python MCP server
3. Local LLMs running through Ollama

This enables:

• Asset Management: Create, import, and manipulate Unity assets programmatically
• Scene Control: Manage scenes, objects, and their properties
• Material Editing: Modify materials and their properties
• Script Integration: View, create, and update Unity scripts
• Editor Automation: Control Unity Editor functions like undo, redo, play, and build

All powered by your own local LLMs, with no need for an internet connection or API keys.

This implementation is specifically configured to work with the following Ollama models:

• deepseek-r1:14b - A 14 billion parameter model with strong reasoning capabilities
• gemma3:12b - Google's 12 billion parameter model with good general capabilities

You can easily switch between these models in the Unity MCP window.

Due to Unity's package manager compatibility issues, we recommend using the Asset Method for installation.

• Unity 2020.3 LTS or newer
• Python 3.10 or newer
• Ollama installed on your system
• The following LLM models pulled in Ollama:
  • ollama pull deepseek-r1:14b
  • ollama pull gemma3:12b

1. Download or clone this repository:

   git clone https://github.com/ZundamonnoVRChatkaisetu/unity-mcp-ollama.git
   

2. Create a folder in your Unity project's Assets directory:

   Assets/UnityMCPOllama
   

3. Copy the Editor folder from the cloned repository to your Unity project:

   # Copy the entire Editor folder
   [Repository]/Editor → Assets/UnityMCPOllama/Editor
   

4. Verify the folder structure is correct:

   Assets/
     UnityMCPOllama/
       Editor/
         MCPEditorWindow.cs
         UnityMCPBridge.cs
   

5. Let Unity import and compile the scripts

1. Create a folder for the Python environment (outside your Unity project):

   mkdir PythonMCP
   cd PythonMCP
   

2. Copy the Python folder from the cloned repository:

   cp -r [Repository]/Python .
   

3. Create and activate a virtual environment:

   # Create a virtual environment
   python -m venv venv
   
   # Activate the virtual environment
   # On Windows:
   venv\Scripts\activate
   # On macOS/Linux:
   source venv/bin/activate

4. Install dependencies:

   cd Python
   pip install -e .

1. Ensure Ollama is installed and running on your system
2. Pull the supported models:

   ollama pull deepseek-r1:14b
   ollama pull gemma3:12b

3. Start Ollama server:

   ollama serve

1. Open your Unity project
2. Navigate to Window > Unity MCP to open the MCP window
3. Click the Start Bridge button to start the Unity bridge

1. Open a command prompt or terminal
2. Navigate to your Python environment:

   cd PythonMCP

3. Activate the virtual environment:

   # On Windows:
   venv\Scripts\activate
   # On macOS/Linux:
   source venv/bin/activate

4. Navigate to the Python directory and start the server:

   cd Python
   python server.py

1. In the Unity MCP window, locate the Ollama Configuration section
2. Verify or update the following settings:
   • Host: localhost (default)
   • Port: 11434 (default)
   • Model: Select either deepseek-r1:14b or gemma3:12b
   • Temperature: Adjust as needed (0.0-1.0)
3. Click Apply Ollama Configuration

1. Click the Show Chat Interface button in the Unity MCP window
2. Type your instructions in the message field
3. Click Send to process your request

Example prompts:

• "Create a red cube at position (0, 1, 0)"
• "Add a sphere to the scene and apply a blue material"
• "List all objects in the current scene"
• "Write a simple movement script and attach it to the cube"

The Unity MCP window provides status information for each component:

• Python Server Status: Indicates whether the Python server is running

  • Green: Connected
  • Yellow: Connected but with issues
  • Red: Not connected

• Unity Bridge Status: Shows if the Unity socket server is running

  • Running: Unity is listening for connections
  • Stopped: Unity socket server is not active

• Ollama Status: Shows the connection status to Ollama

  • Connected: Successfully connected to Ollama server
  • Not Connected: Unable to connect to Ollama

1. "Not Connected" Status for Python Server

   • Ensure the Python server is running (python server.py)
   • Check for errors in the Python console
   • Verify the Unity Bridge is running

2. Cannot find Unity MCP menu

   • Make sure the Editor scripts are properly imported in your project
   • Check the Unity console for any errors
   • Restart Unity if necessary

3. Ollama Connection Issues

   • Verify Ollama is running with ollama serve
   • Check that models are properly pulled
   • Ensure no firewall is blocking port 11434

4. MCP Command Execution Fails

   • Check Python console for detailed error messages
   • Verify that the Unity Bridge is running
   • Make sure the prompt is clear and specific

If you encounter issues setting up the Python environment:

1. Install Python 3.10 or newer
2. Install Ollama from ollama.ai
3. Create a dedicated directory for the Python environment:

   mkdir C:\PythonMCP
   cd C:\PythonMCP
   

4. Clone or download this repository and copy the Python folder:

   git clone https://github.com/ZundamonnoVRChatkaisetu/unity-mcp-ollama.git
   copy unity-mcp-ollama\Python .
   

5. Create a virtual environment:

   python -m venv venv
   

6. Activate the virtual environment:

   venv\Scripts\activate
   

7. Install dependencies:

   cd Python
   pip install -e .
   

8. Run the server:

   python server.py
   

Local LLM performance depends on your hardware:

• For deepseek-r1:14b: Recommended minimum 12GB VRAM
• For gemma3:12b: Recommended minimum 10GB VRAM
• CPU-only operation is possible but will be significantly slower

Contributions are welcome! Please feel free to submit a Pull Request or open an Issue.

This project is licensed under the MIT License.

• Based on justinpbarnett/unity-mcp
• Uses Ollama for local LLM integration