A Quick Look: How it Connects AI to Game Engines

GameCatalyst uses the Model Context Protocol (MCP), a standardized way for AI assistants to communicate with tools. When you connect an AI client like Claude Desktop or ChatGPT, GameCatalyst appears as a set of tools the AI can use.

Here’s the flow:

1. You connect your AI client — You add GameCatalyst to your AI client’s configuration file or web interface.
2. GameCatalyst registers your Unity project — The system detects your open Unity project and makes it available.
3. You select the project — You explicitly tell your AI which project to work on. The AI never auto-selects.
4. Commands become available — Your AI can now see 60+ commands for creating objects, modifying scenes, managing files, and more.
5. You give instructions — You ask the AI to do something in natural language.
6. AI executes commands — The AI translates your request into GameCatalyst commands and executes them.
7. Results appear in Unity — Changes happen in real time inside your Unity Editor.

GameCatalyst handles all the technical complexity: session management, permission checks, command validation, and thread-safe execution. You just talk to your AI naturally, and things happen.

The key insight:

GameCatalyst uses the Model Context Protocol (MCP), which is a standardized way for AI tools to communicate. Think of MCP like USB — it is a universal standard that any AI client can implement. When a new AI tool launches and it supports MCP, it automatically works with GameCatalyst without any changes on your end.

What happens behind the scenes:

GameCatalyst runs several services on your local machine: a Discovery Server that tracks your projects, an MCP Bridge that handles AI communication, and an Engine Instance Server that connects to Unity. These services run as separate processes outside Unity, which means they stay active even when Unity recompiles your code. All the technical complexity is hidden from you — you just talk to your AI naturally and things happen in Unity.

Security and control:

The AI never selects a project automatically. You must explicitly choose which project to work on. You control which commands the AI can use. File write operations simulate by default (dry-run mode) so you see what would happen before it actually happens. Every layer is designed to keep you in control.