.agent/skills/unity-initial-setup/SKILL.md
Provides an initial setup for AI Skills, `unity-mcp-cli` command line tool installation and everything else that is helpful to set up at the start of the project. Essential packages, and basic configurations.
npx skillsauth add hth-hiro/Unity-Study unity-initial-setupInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
This guide walks through installing the unity-mcp-cli command-line tool and using it to set
up a Unity project with AI Skills and MCP integration.
unity-mcp-cli requires Node.js ^20.19.0 || >=22.12.0 (Node 21.x is not supported). If you don't have Node.js installed:
Download the installer from https://nodejs.org/ and run it, or use a package manager:
winget install OpenJS.NodeJS.LTS
After installation, verify both node and npm are available:
node --version
npm --version
unity-mcp-cli GloballyInstall the CLI globally so the unity-mcp-cli command is available system-wide:
npm install -g unity-mcp-cli
Verify installation:
unity-mcp-cli --version
Alternative: Run any command without installing globally using
npx:npx unity-mcp-cli --help
Add the plugin to an existing Unity project:
unity-mcp-cli install-plugin /path/to/unity/project
Enable all tools, prompts, and resources:
unity-mcp-cli configure /path/to/unity/project \
--enable-all-tools \
--enable-all-prompts \
--enable-all-resources
Configure MCP integration for your AI agent (e.g., Claude Code, Cursor, Copilot):
unity-mcp-cli setup-mcp claude-code /path/to/unity/project
List all supported agents:
unity-mcp-cli setup-mcp --list
Generate skill files for your AI agent (requires Unity Editor to be running with the plugin):
unity-mcp-cli setup-skills claude-code /path/to/unity/project
unity-mcp-cli open /path/to/unity/project
| Command | Description |
|---|---|
| unity-mcp-cli install-plugin <path> | Install Unity-MCP plugin into a project |
| unity-mcp-cli remove-plugin <path> | Remove Unity-MCP plugin from a project |
| unity-mcp-cli configure <path> --list | List current MCP configuration |
| unity-mcp-cli setup-mcp <agent> <path> | Write MCP config for an AI agent |
| unity-mcp-cli setup-skills <agent> <path> | Generate skill files for an AI agent |
| unity-mcp-cli create-project <path> | Create a new Unity project |
| unity-mcp-cli install-unity <version> | Install a Unity Editor version |
| unity-mcp-cli open <path> | Open a Unity project in the Editor |
| unity-mcp-cli run-tool <tool> <path> | Execute an MCP tool via HTTP API |
Add --verbose to any command for detailed diagnostic output.
npm not found: Node.js is not installed or not in your PATH. Reinstall Node.js and restart your terminal.install-plugin, open the project in Unity Editor. The package manager resolves dependencies on project open.setup-skills.tools
Generate all skills from the existed Tools in the Unity Project.
tools
Create a new skill using C# code. It will be added into the project as a .cs file and compiled by Unity. The skill will be available for use after compilation. It must be a partial class decorated with [McpPluginToolType]. Each tool method must be decorated with [McpPluginTool]. The class name should match the file name. All Unity API calls must use com.IvanMurzak.ReflectorNet.Utils.MainThread.Instance.Run(). Return a data model for structured output, or void for side-effect-only operations. Full sample: ```csharp #nullable enable using System; using System.ComponentModel; using com.IvanMurzak.McpPlugin; using com.IvanMurzak.ReflectorNet.Utils; using com.IvanMurzak.Unity.MCP.Editor.Utils; using com.IvanMurzak.Unity.MCP.Runtime.Data; using UnityEditor; using UnityEngine; namespace com.IvanMurzak.Unity.MCP.Editor.API { [McpPluginToolType] public partial class Tool_Sample { [McpPluginTool("sample-get", Title = "Sample / Get")] [Description("Finds a GameObject and returns its ref data.")] public GameObjectRef Get ( [Description("Name of the GameObject to find.")] string name ) { return MainThread.Instance.Run(() => { var go = GameObject.Find(name) ?? throw new ArgumentException($"GameObject '{name}' not found.", nameof(name)); return new GameObjectRef(go); }); } [McpPluginTool("sample-rename", Title = "Sample / Rename")] [Description("Renames a GameObject.")] public void Rename ( [Description("Current name of the GameObject.")] string name, [Description("New name to assign.")] string newName ) { MainThread.Instance.Run(() => { var go = GameObject.Find(name) ?? throw new ArgumentException($"GameObject '{name}' not found.", nameof(name)); go.name = newName; EditorUtility.SetDirty(go); AssetDatabase.Refresh(ImportAssetOptions.ForceSynchronousImport); EditorUtils.RepaintAllEditorWindows(); }); } } } ``` ## Suggestions ### Refresh UI after visual changes If the skill modifies anything visually in the Unity Editor (GameObjects, components, materials, etc.), call these two lines at the end of the tool method to apply changes to the UI immediately: ```csharp AssetDatabase.Refresh(ImportAssetOptions.ForceSynchronousImport); EditorUtils.RepaintAllEditorWindows(); ``` ### Refresh AssetDatabase after asset or script changes If the skill creates, modifies, or deletes any asset file or .cs script on disk outside of Unity API, call this inside a `MainThread.Instance.Run()` block to ensure Unity picks up the changes: ```csharp MainThread.Instance.Run(() => { AssetDatabase.Refresh(ImportAssetOptions.ForceSynchronousImport); }); ``` ### Use processing mechanic for long-running or domain-reload operations Some operations take time to complete and may trigger a Unity domain reload (e.g. writing a .cs script, switching play mode, running tests, adding a package). In these cases the tool must NOT block and wait — instead it must: 1. Accept a `[RequestID] string? requestId` parameter. 2. Return `ResponseCallTool.Processing("...").SetRequestID(requestId)` immediately. 3. Schedule the actual work asynchronously via `MainThread.Instance.RunAsync(async () => { await Task.Yield(); ... })`. 4. When the operation finishes, send the final result by calling: ```csharp _ = UnityMcpPluginEditor.NotifyToolRequestCompleted(new RequestToolCompletedData { RequestId = requestId, Result = ResponseCallTool.Success("Operation completed.").SetRequestID(requestId) }); ``` If the operation may survive a domain reload (e.g. a .cs file was saved and Unity will recompile), use `ScriptUtils.SchedulePostCompilationNotification(requestId, filePath, operationType)` instead of calling `NotifyToolRequestCompleted` directly — it persists the pending notification to `SessionState` and sends it automatically after the domain reload completes. For package install/removal or other non-compilation domain reloads use `PackageUtils.SchedulePostDomainReloadNotification(requestId, label, action, expectedResult)` the same way. ### Return structured data with a typed response Prefer returning a structured data model over a plain string so the AI can parse individual fields. Declare a nested class with `[Description]` on each property and use `ResponseCallValueTool<T>` as return type: ```csharp // Return type: public ResponseCallValueTool<MyResult> MyTool(...) { return ResponseCallValueTool<MyResult>.Success(new MyResult { Name = go.name, InstanceID = go.GetInstanceID() }).SetRequestID(requestId); } // Data model: public class MyResult { [Description("Name of the GameObject.")] public string? Name { get; set; } [Description("Unity instance ID of the GameObject.")] public int InstanceID { get; set; } } ``` For simpler cases that do not need async/processing, you may return the model directly (without `ResponseCallValueTool<T>`) and Unity-MCP will wrap it automatically. ### Validate inputs early and throw clearly Always validate required parameters at the top of the method before any Unity API calls. Throw `ArgumentException` or `InvalidOperationException` with descriptive messages so the AI knows exactly what went wrong and can self-correct: ```csharp if (string.IsNullOrEmpty(name)) throw new ArgumentException("Name cannot be null or empty.", nameof(name)); ``` ### Always use MainThread for Unity API calls All Unity API calls (including `GameObject.Find`, `AssetDatabase`, `EditorUtility`, etc.) MUST run on the main thread. Wrap them in `MainThread.Instance.Run(() => { ... })` for synchronous operations, or `MainThread.Instance.RunAsync(async () => { ... })` when you need to await inside.
tools
Provides an initial setup for AI Skills, `unity-mcp-cli` command line tool installation and everything else that is helpful to set up at the start of the project. Essential packages, and basic configurations.
data-ai
Generates a JSON Schema for a given C# type name using reflection. Supports primitives, enums, arrays, generic collections, dictionaries, and complex objects. The type must be present in any loaded assembly. Use the full type name (e.g. 'UnityEngine.Vector3') for best results.