Unity MCP servers compared
There are several good ways to connect an AI to Unity. The biggest practical difference is what you have to install and run besides Unity. Checked against each project’s own documentation on 2026-10-19; things change, so check the projects themselves before deciding.
| Needs besides Unity | Unity versions | |
|---|---|---|
| nanora-unity-mcp (MIT) | Nothing — C# inside the Editor process; Roslyn is bundled | 2021.3 – 6 |
| CoplayDev/unity-mcp (MIT) | Python 3.10+ (uv) | 2021.3 LTS – 6.x |
| CoderGamester/mcp-unity (MIT) | Node.js 18+; a WebSocket on the Unity side | 6 or later |
| IvanMurzak/Unity-MCP (Apache-2.0) | A separate server executable | — |
| Unity’s own MCP (AI Assistant) | A relay program, a project linked to Unity Cloud, and an AI trial or subscription | 6 or later |
Where nanora-unity-mcp fits
Section titled “Where nanora-unity-mcp fits”- Only Unity. No Python, Node.js, extra executable or relay.
- Streamable HTTP and SSE on one server, so Codex and Claude Code or Cursor can use the same Editor.
- A fixed port per project, a written client config and a
/healththat names the project, so several Editors do not get mixed up without any setup. - Checking, not only editing. U-Selector with
wait_for_selector,assert_ui_stateandsimulate_input, and skills that make the AI verify its work.
Where others are ahead
Section titled “Where others are ahead”- Other servers have more tools (IvanMurzak’s has 70+ with extensions) and larger communities.
- Remote use (OAuth, cloud relays) is not a goal here: nanora-unity-mcp is local only.
- Running C# and handling several Editors are available elsewhere too.