Skip to main content

Ollama

This page explains how to connect OpenForge MCP with Ollama.

What Is Ollama?​

Ollama is a tool that lets you run AI (large language models) from the terminal. You can download and run models with a single command, and everything runs locally. There are no API fees, and it works even without an internet connection.

When connected to OpenForge MCP, you can control 3D applications like Unity and Blender through terminal conversations.

Note

Ollama is a terminal-based tool. If you prefer a graphical interface, LM Studio might be easier to use.

Prerequisites​

Installing Ollama​

Download the installer from ollama.com and install it.

Once the installation is complete, run the following command in your terminal to verify:

ollama --version

Downloading a Model​

To use AI with Ollama, you first need to download a model.

ollama pull llama3

This command downloads the "llama3" model. Depending on the model size, the download may be several GB and take some time.

Model NameSizeFeatures
llama38BWell balanced, recommended for beginners
codellama7BStrong at code generation. Ideal when combined with script creation
mistral7BLightweight and fast. Good when response speed is a priority

For higher performance, you can also use larger models such as llama3:70b or qwen2.5:72b (24 GB or more of GPU memory recommended).

Setup​

Automatic Setup​

Run the following command in your terminal to automatically configure Essential Mode.

npx openforge-mcp setup --client ollama --mode essential

If you see "Setup complete", you are done.

Manual Setup​

To use OpenForge MCP from Ollama, you connect through the HTTP API.

Step 1: Start the OpenForge MCP Server​

Run the following command in your terminal to start OpenForge MCP in HTTP server mode.

npx openforge-mcp start --transport http --mode essential

When the server starts, you will see the endpoint URL:

OpenForge MCP HTTP server listening on http://localhost:24601
Note

Do not close this terminal window. The server needs to keep running.

Step 2: Start a Model with Ollama​

Open another terminal window and start a model with Ollama.

ollama run llama3

The interactive mode will begin.

About Essential Mode​

Local AI models have more limited context compared to cloud AI. In Essential Mode, only about 80 of the most commonly used tools from OpenForge MCP are loaded.

This allows comfortable operation within the limited context of local models. All basic 3D operations (creating objects, moving them, deleting them, setting materials, and so on) are covered.

Usage​

With the OpenForge MCP server running, give instructions from Ollama's interactive mode.

Usage Examples​

$ ollama run llama3

> Create a Cube in the Unity scene

Created a Cube.

> Change the position to 0, 2, 0

Changed the Cube's position to (0, 2, 0).

> Show me the list of objects in the scene

The current scene contains the following objects:
- Main Camera
- Directional Light
- Cube

Verification​

With the OpenForge MCP server running, execute the following command from another terminal:

npx openforge-mcp test

If you see "All tests passed", OpenForge MCP is working correctly.

Next, try typing the following in Ollama's interactive mode:

"Show me the OpenForge MCP tool list"

If tool information is returned, the connection is working.

Troubleshooting​

"Connection refused" Error​

  • Check that the OpenForge MCP HTTP server is running
  • Look for errors in the terminal where the server was started
  • Verify the port number (default: 24601)

Cannot Download a Model with Ollama​

  • Check your internet connection (only needed for downloading models)
  • Check your available disk space. Models can be several GB in size
  • Run ollama list to see if you already have downloaded models

The AI Does Not Use Tools Correctly​

  • Try a larger model (13B or above). Smaller models may be less accurate at tool calling
  • Simplify your instructions. Giving one operation at a time is the key
  • Confirm that Essential Mode is enabled

Slow Performance​

  • Check whether the GPU is available. Ollama auto-detects the GPU, but an outdated driver may cause it to fall back to CPU
  • Try a smaller model
  • Run ollama ps to see currently running models. If unneeded models are running, stop them with ollama stop modelname

npx Command Not Found​

Node.js may not be installed. Download and install the LTS version from nodejs.org.

That completes the Ollama setup. If you encounter any issues, see the FAQ as well.