mcp-unity MCP工具

Requirements

• Unity 6 or later - to install the server
• Node.js 18 or later - to start the server
• npm 9 or later - to debug the server

[!NOTE] Project Paths with Spaces MCP Unity supports project paths containing spaces. However, if you experience connection issues, try moving your project to a path without spaces as a troubleshooting step. Examples: - ✅ Recommended: C:\Users\YourUser\Documents\UnityProjects\MyAwesomeGame - ✅ Supported: C:\Users\Your User\Documents\Unity Projects\My Awesome Game

<a name="install-server"></a>Installation

Installing this MCP Unity Server is a multi-step process:

Step 1: Install Node.js

To run MCP Unity server, you'll need to have Node.js 18 or later installed on your computer:

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">Windows</span></summary>

1. Visit the Node.js download page 2. Download the Windows Installer (.msi) for the LTS version (recommended) 3. Run the installer and follow the installation wizard 4. Verify the installation by opening PowerShell and running:

   node --version
   

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">macOS</span></summary>

1. Visit the Node.js download page 2. Download the macOS Installer (.pkg) for the LTS version (recommended) 3. Run the installer and follow the installation wizard 4. Alternatively, if you have Homebrew installed, you can run:

   brew install node@18
   

   node --version
   

Step 2: Install Unity MCP Server package via Unity Package Manager

1. Open the Unity Package Manager (Window > Package Manager)
2. Click the "+" button in the top-left corner
3. Select "Add package from git URL..."
4. Enter: https://github.com/CoderGamester/mcp-unity.git
5. Click "Add"

Step 3: Configure AI LLM Client

<details open> <summary><span style="font-size: 1.1em; font-weight: bold;">Option 1: Configure using Unity Editor</span></summary>

1. Open the Unity Editor
2. Navigate to Tools > MCP Unity > Server Window
3. Click on the "Configure" button for your AI LLM client as shown in the image below

Global vs. Project configuration: > - Configure \[Client\] — writes to your global user config file (e.g. ~/.claude.json). Uses an absolute path. Applies to all projects on your machine. Best for personal, single-developer setups. > - Configure \[Client\] (Project) — writes to a .mcp.json file (or equivalent) in the Unity project root. Uses a relative path, so it works across machines. Intended to be committed to git and shared with the team. Best for collaborative projects or when you want the config to travel with the project. > > If in doubt, prefer the (Project) variant — the relative path is more portable and won't break if you move your project folder.

1. Confirm the configuration installation with the given popup

</details>

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">Option 2: Configure Manually</span></summary>

Open the MCP configuration file of your AI client and add the MCP Unity server configuration:

Replace ABSOLUTE/PATH/TO with the absolute path to your MCP Unity installation or just copy the text from the Unity Editor MCP Server window (Tools > MCP Unity > Server Window). For configs that live inside the Unity project tree and get committed to git (e.g. <project>/.vscode/mcp.json, <project>/opencode.json, <project>/.cursor/mcp.json, <project>/.mcp.json, <project>/.codex/config.toml), prefer a project-relative path so the same file works across machines. Toggle "Use relative path" in the Server Window to switch the copy-paste snippet between absolute and project-relative forms. The Configure GitHub Copilot, Configure OpenCode, Configure Cursor (Project), Configure Claude Code (Project), and Configure Codex CLI (Project) buttons already emit relative paths automatically. Project-local buttons (Cursor / Claude Code / Codex CLI) write the MCP server entry into the Unity project directory instead of your global user config, so other (non-Unity) projects don't see MCP connection-failure warnings. For Codex CLI (Project) specifically, you must approve the project trust prompt the first time you run codex from the project root, otherwise Codex ignores <project>/.codex/config.toml.

For JSON-based clients (Cursor, Windsurf, Claude Code, GitHub Copilot, etc.):

{
   "mcpServers": {
       "mcp-unity": {
          "command": "node",
          "args": [
             "ABSOLUTE/PATH/TO/mcp-unity/Server~/build/index.js"
          ]
       }
   }
}

For workspace-scoped VS Code / GitHub Copilot (.vscode/mcp.json), use ${workspaceFolder} so the path is portable across machines:

{
   "mcpServers": {
       "mcp-unity": {
          "command": "node",
          "args": [
             "${workspaceFolder}/Library/PackageCache/com.gamelovers.mcp-unity@<hash>/Server~/build/index.js"
          ]
       }
   }
}

For Codex CLI (~/.codex/config.toml):

[mcp_servers.mcp-unity]
command = "node"
args = ["ABSOLUTE/PATH/TO/mcp-unity/Server~/build/index.js"]

For Cursor — project-local (.cursor/mcp.json in the Unity project root, project-relative path):

{
   "mcpServers": {
       "mcp-unity": {
          "command": "node",
          "args": [
             "Library/PackageCache/com.gamelovers.mcp-unity@<hash>/Server~/build/index.js"
          ]
       }
   }
}

For Claude Code — project-local (.mcp.json in the Unity project root, project-relative path — Claude Code's team-shared MCP config):

{
   "mcpServers": {
       "mcp-unity": {
          "command": "node",
          "args": [
             "Library/PackageCache/com.gamelovers.mcp-unity@<hash>/Server~/build/index.js"
          ]
       }
   }
}

For Codex CLI — project-local (.codex/config.toml in the Unity project root, project-relative path):

[mcp_servers.mcp-unity]
command = "node"
args = ["Library/PackageCache/com.gamelovers.mcp-unity@<hash>/Server~/build/index.js"]

Codex layers this file over the global ~/.codex/config.toml, but only when the project is marked trusted. The first time you cd into the project and run codex, approve the trust prompt — otherwise Codex ignores .codex/config.toml.

For OpenCode (opencode.json in the Unity project root):

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "mcp-unity": {
      "type": "local",
      "enabled": true,
      "command": ["node", "Library/PackageCache/com.gamelovers.mcp-unity@<hash>/Server~/build/index.js"],
      "environment": {}
    }
  }
}

Note: the @<hash> segment in the UPM package cache path changes when the package is updated. If you update MCP Unity, re-run the Configure button (or update the path manually) so the snippet points at the new cache directory.

</details>

Optional: Set WebSocket Port

By default, the WebSocket server runs on port '8090'. You can change this port in two ways:

1. Open the Unity Editor
2. Navigate to Tools > MCP Unity > Server Window
3. Change the "WebSocket Port" value to your desired port number
4. Unity will setup the system environment variable UNITY_PORT to the new port number
5. Restart the Node.js server
6. Click again on "Start Server" to reconnect the Unity Editor web socket to the Node.js MCP Server

Optional: Set Timeout

By default, the timeout between the MCP server and the WebSocket is 10 seconds. You can change depending on the OS you are using:

1. Open the Unity Editor
2. Navigate to Tools > MCP Unity > Server Window
3. Change the "Request Timeout (seconds)" value to your desired timeout seconds
4. Unity will setup the system environment variable UNITY_REQUEST_TIMEOUT to the new timeout value
5. Restart the Node.js server
6. Click again on "Start Server" to reconnect the Unity Editor web socket to the Node.js MCP Server

[!TIP] The timeout between your AI Coding IDE (e.g., Claude Desktop, Cursor IDE, Windsurf IDE) and the MCP Server depends on the IDE.

Optional: Allow Remote MCP Bridge Connections

By default, the WebSocket server binds to 'localhost'. To allow MCP bridge connections from other machines:

1. Open the Unity Editor
2. Navigate to Tools > MCP Unity > Server Window
3. Enable the "Allow Remote Connections" checkbox
4. Unity will bind the WebSocket server to '0.0.0.0' (all interfaces)
5. Restart the Node.js server to apply the new host configuration
6. Set the environment variable UNITY_HOST to your Unity machine's IP address when running the MCP bridge remotely: UNITY_HOST=192.168.1.100 node server.js

IDE Integration - Package Cache Access

MCP Unity provides automatic integration with VSCode-like IDEs (Visual Studio Code, Cursor, Windsurf, Google Antigravity) by adding the Unity Library/PackedCache folder to your workspace. This feature:

• Improves code intelligence for Unity packages
• Enables better autocompletion and type information for Unity packages
• Helps AI coding assistants understand your project's dependencies

Frequently Asked Questions

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">What is MCP Unity?</span></summary>

MCP Unity is a powerful bridge that connects your Unity Editor environment to AI assistants LLM tools using the Model Context Protocol (MCP).

In essence, MCP Unity: - Exposes Unity Editor functionalities (like creating objects, modifying components, running tests, etc.) as "tools" and "resources" that an AI can understand and use. - Runs a WebSocket server inside Unity and a Node.js server (acting as a WebSocket client to Unity) that implements the MCP. This allows AI assistants to send commands to Unity and receive information back. - Enables you to use natural language prompts with your AI assistant to perform complex tasks within your Unity project, significantly speeding up development workflows.

</details>

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">Why use MCP Unity?</span></summary>

MCP Unity offers several compelling advantages for developers, artists, and project managers:

• Accelerated Development: Automate repetitive tasks, generate boilerplate code, and manage assets using AI prompts. This frees up your time to focus on creative and complex problem-solving.
• Enhanced Productivity: Interact with Unity Editor features without needing to manually click through menus or write scripts for simple operations. Your AI assistant becomes a direct extension of your capabilities within Unity.
• Improved Accessibility: Allows users who are less familiar with the deep intricacies of the Unity Editor or C# scripting to still make meaningful contributions and modifications to a project through AI guidance.
• Seamless Integration: Designed to work with various AI assistants and IDEs that support MCP, providing a consistent way to leverage AI across your development toolkit.
• Extensibility: The protocol and the toolset can be expanded. You can define new tools and resources to expose more of your project-specific or Unity's functionality to AI.
• Collaborative Potential: Facilitates a new way of collaborating where AI can assist in tasks traditionally done by team members, or help in onboarding new developers by guiding them through project structures and operations.

</details>

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">How does MCP Unity compare with the upcoming Unity 6.2 AI features?</span></summary>

Unity 6.2 is set to introduce new built-in AI tools, including the previous Unity Muse (for generative AI capabilities like texture and animation generation) and Unity Sentis (for running neural networks in Unity runtime). As Unity 6.2 is not yet fully released, this comparison is based on publicly available information and anticipated functionalities:

• Focus:
• MCP Unity: Primarily focuses on Editor automation and interaction. It allows external AI (like LLM-based coding assistants) to control and query the Unity Editor itself to manipulate scenes, assets, and project settings. It's about augmenting the developer's workflow within the Editor.
• Unity 6.2 AI:
• Aims at in-Editor content creation (generating textures, sprites, animations, behaviors, scripts) and AI-powered assistance for common tasks, directly integrated into the Unity Editor interface.
• A fine-tuned model to ask any question about Unity's documentation and API structure, with customized examples more accurate to Unity's environment.
• Adds the functionality to run AI model inference, allowing developers to deploy and run pre-trained neural networks within your game or application for features like NPC behavior, image recognition, etc.

• Use Cases:
• MCP Unity: "Create a new 3D object, name it 'Player', add a Rigidbody, and set its mass to 10." "Run all Play Mode tests." "Ask to fix the error on the console log." "Execute the custom menu item 'Prepare build for iOS' and fix any errors that may occur."
• Unity 6.2 AI: "Generate a sci-fi texture for this material." "Update all trees position in the scene to be placed inside of terrain zones tagged with 'forest'." "Create a walking animation for this character." "Generate 2D sprites to complete the character." "Ask details about the error on the console log."

- Complementary, Not Mutually Exclusive: MCP Unity and Unity's native AI tools can be seen as complementary. You might use MCP Unity with your AI coding assistant to set up a scene or batch-modify assets, and then use Unity AI tools to generate a specific texture, or to create animations, or 2D sprites for one of those assets. MCP Unity provides a flexible, protocol-based way to interact with the Editor, which can be powerful for developers who want to integrate with a broader range of external AI services or build custom automation workflows.

</details>

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">What MCP hosts and IDEs currently support MCP Unity?</span></summary>

MCP Unity is designed to work with any AI assistant or development environment that can act as an MCP client. The ecosystem is growing, but current known integrations or compatible platforms include: - Cursor - Windsurf - Claude Desktop - Claude Code - Codex CLI - GitHub Copilot - Google Antigravity - OpenCode

</details>

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">Can I extend MCP Unity with custom tools for my project?</span></summary>

Yes, absolutely! One of the significant benefits of the MCP Unity architecture is its extensibility. - In Unity (C#): You can create new C# classes that inherit from McpToolBase (or a similar base for resources) to expose custom Unity Editor functionality. These tools would then be registered in McpUnityServer.cs. For example, you could write a tool to automate a specific asset import pipeline unique to your project. - In Node.js Server (TypeScript): You would then define the corresponding TypeScript tool handler in the Server/src/tools/ directory, including its Zod schema for inputs/outputs, and register it in Server/src/index.ts. This Node.js part will forward the request to your new C# tool in Unity.

This allows you to tailor the AI's capabilities to the specific needs and workflows of your game or application.

</details>

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">Is MCP Unity free to use?</span></summary>

Yes, MCP Unity is an open-source project distributed under the MIT License. You are free to use, modify, and distribute it according to the license terms.

</details>

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">Why am I unable to connect to MCP Unity?</span></summary>

• Ensure the WebSocket server is running (check the Server Window in Unity)
• Send a console log message from MCP client to force a reconnection between MCP client and Unity server
• Change the port number in the Unity Editor MCP Server window. (Tools > MCP Unity > Server Window)

</details>

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">Why won't the MCP Unity server start?</span></summary>

• Check the Unity Console for error messages
• Ensure Node.js is properly installed and accessible in your PATH
• Verify that all dependencies are installed in the Server directory

</details>

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">Why do I get a connection failed error when running Play Mode tests?</span></summary>

The run_tests tool returns the following response:

Error:
Connection failed: Unknown error

This error occurs because the bridge connection is lost when the domain reloads upon switching to Play Mode. The workaround is to turn off Reload Domain in Edit > Project Settings > Editor > "Enter Play Mode Settings".

</details>

<details> <summary><span style="font-size: 1.1em; font-weight: bold;">Why do some clients fail with <code>KeyError: 'position'</code> during tool initialization?</span></summary>

Some MCP clients may fail while parsing tool schemas when they contain local JSON pointer references such as #/properties/position.

MCP Unity avoids this by registering transform tool inputs (set_transform, move_gameobject, rotate_gameobject, scale_gameobject) with fresh nested vector schemas per field, so the generated schema does not rely on local #/properties/... references.

If you still see this error: - update your MCP client to the latest version, - rebuild the Node server (cd Server~ && npm run build), - confirm your package version includes this compatibility fix.

</details>

Troubleshooting: WSL2 (Windows 11) networking

When running the MCP (Node.js) server inside WSL2 while Unity runs on Windows 11, connecting to ws://localhost:8090/McpUnity may fail with ECONNREFUSED.

Cause: WSL2 and Windows have separate network namespaces — localhost inside WSL2 does not point to the Windows host. By default, Unity listens on localhost:8090.