# Unity MCP — Control Unity Editor from Agents

> Unity MCP bridges MCP clients to your Unity Editor so assistants can manage scenes, assets, scripts, and builds via tools instead of manual clicking.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

1. In Unity Package Manager: “Add package from git URL…”
   - `https://github.com/CoplayDev/unity-mcp.git?path=/MCPForUnity#beta`
2. Open: `Window > MCP for Unity` and run the setup.
3. Add the HTTP config (default):
   ```json
   { "mcpServers": { "unityMCP": { "url": "http://localhost:8080/mcp" } } }
   ```
4. Ask your client to list `unity_instances` and try one tool call.

## Intro

Unity MCP bridges MCP clients to your Unity Editor so assistants can manage scenes, assets, scripts, and builds via tools instead of manual clicking.

- **Best for:** Unity teams who want LLMs to automate editor workflows (scene edits, package management, build pipelines)
- **Works with:** Unity 2021.3+ editor, Python 3.10+, MCP clients (Claude Desktop/Code, Cursor, VS Code, etc.)
- **Setup time:** 20 minutes

## Practical Notes

- Release notes in README show versioned tool growth (for example v9.6.1 adds build tooling and multi-scene ops)
- Performance tip: README claims `batch_execute` can be 10–100× faster than individual calls

## Treat Unity as an API, Not a GUI

The win is not “an LLM can click buttons” — it’s that your editor workflow becomes a **repeatable tool contract**:

- Use `batch_execute` for multi-step operations so you get one atomic-ish plan and fewer round-trips.
- Route requests to the correct Unity instance when you have multiple editors open.
- Keep transports local-first (loopback) unless you truly need LAN binding.

Once connected, you can standardize tasks:

- “Create a new scene from template, add a prefab, set lighting defaults”
- “Scan packages, add a scoped registry, then validate compile”
- “Build a player with a known profile and export artifacts”

That makes AI-assisted Unity work auditable: every change is a tool call you can reproduce.

### FAQ

**Q: Do I need Python?**
A: Yes. The README lists Python 3.10+ and uv as requirements for the bridge/server pieces.

**Q: Which transport should I start with?**
A: HTTP localhost is the default and works with many clients; stdio is also available.

**Q: How do I speed up multi-step edits?**
A: Use `batch_execute` (README claims 10–100× faster).

## Source & Thanks

> Source: https://github.com/CoplayDev/unity-mcp
> License: MIT
> GitHub stars: 9,486 · forks: 1,074

---

<!-- ZH -->

## 快速使用

1. Unity Package Manager：Add package from git URL…
   - `https://github.com/CoplayDev/unity-mcp.git?path=/MCPForUnity#beta`
2. 打开：`Window > MCP for Unity` 并完成一键 setup。
3. 默认 HTTP 配置：
   ```json
   { "mcpServers": { "unityMCP": { "url": "http://localhost:8080/mcp" } } }
   ```
4. 让客户端读取 `unity_instances`，然后尝试调用一个 tool。

## 简介

Unity MCP 把 Unity Editor 变成 MCP 工具端：让助手通过 tools 管理场景/资源/脚本/构建，把重复点点点变成可复用指令，并把每次编辑沉淀为可追溯的工具调用记录。

- **适合谁：** 希望用 LLM 自动化 Unity 编辑器流程（改场景、管包、跑构建）的团队
- **可搭配：** Unity 2021.3+、Python 3.10+、MCP 客户端（Claude Desktop/Code、Cursor、VS Code 等）
- **准备时间：** 20 分钟

## 实战建议

- README 的版本记录显示工具集持续扩展（例如 v9.6.1 增加 build 工具与多场景编辑）
- 性能建议：README 提到 `batch_execute` 比逐个调用快 10–100×

## 把 Unity 当 API，而不是 GUI

真正的收益不是“LLM 会点按钮”，而是把编辑器流程变成**可重复的工具契约**：

- 多步骤操作尽量用 `batch_execute` 聚合，减少来回与状态漂移。
- 多 Unity 实例时明确路由到正确实例（避免改错工程）。
- 传输优先本机回环（loopback），除非必要不要开放到 LAN。

连通后你可以把任务标准化：

- “按模板建新场景 → 放 prefab → 设置光照默认值”
- “扫描包 → 加 scoped registry → 校验编译”
- “用固定 profile 构建 player 并导出产物”

这样 AI 辅助 Unity 才可审计：每次改动都是可复现的 tool call。

### FAQ

**必须装 Python 吗？**
答：需要，README 要求 Python 3.10+（并建议使用 uv）。

**优先用哪种传输？**
答：默认是本机 HTTP（兼容性好）；也支持 stdio。

**多步骤操作怎么更快？**
答：用 `batch_execute`（README 提到可提升 10–100×）。

## 来源与感谢

> Source: https://github.com/CoplayDev/unity-mcp
> License: MIT
> GitHub stars: 9,486 · forks: 1,074


---
Source: https://tokrepo.com/en/workflows/unity-mcp-control-unity-editor-from-agents
Author: MCP Hub