Unity-MCP

Unity-MCP is an AI-driven Unity game development assistant, targeting both Editor and Runtime scenarios. It connects AI agents such as Claude, Cursor, and Windsurf to Unity via MCP (Model Context Protocol) to automate workflows, generate code, and let AI run inside the game. According to the official description, what sets it apart from other tools is that the plugin can work in compiled games, thereby supporting real-time AI debugging and player-AI interaction.

Main capabilities

• AI agents: You can use agents from Anthropic, OpenAI, Microsoft, or other providers; officially stated as having no vendor lock-in.
• Tools: Provides a broad set of default MCP tools for operating the Unity Editor.
• Skills: Generates AI skills based on the operating system, Unity version, and plugins in the project.
• Code and testing: Use AI agents to develop game mechanics and test them.
• Runtime (in-game): Use an LLM directly inside a compiled game to implement dynamic NPC behavior or debugging.
• Debugging support: Let AI debug and fix issues in the project.
• Natural conversation: Chat with AI as if talking to a person.
• Flexible deployment: Supports local (stdio) and remote (http) via configuration.
• Extensible: Custom tools can be created in project code.

Built-in tool categories (70+, ready to use after installation with no additional configuration)

1. Assets: assets-copy, assets-create-folder, assets-delete, assets-find, assets-find-built-in, assets-get-data, assets-material-create, assets-modify, assets-move, assets-prefab-close, assets-prefab-create, assets-prefab-instantiate, assets-prefab-open, assets-prefab-save, assets-refresh, assets-shader-get-data, assets-shader-list-all.
2. Package management (package): package-add, package-list, package-remove, package-search.
3. GameObject and components: gameobject-component-add, gameobject-component-destroy, gameobject-component-get, gameobject-component-list-all, gameobject-component-modify, gameobject-create, gameobject-destroy, gameobject-duplicate, gameobject-find, gameobject-modify, gameobject-set-parent.
4. Objects: object-get-data, object-modify.
5. Scene: scene-create, scene-get-data, scene-list-opened, scene-open, scene-save, scene-set-active, scene-unload.
6. Screenshot: screenshot-camera, screenshot-game-view, screenshot-isolated, screenshot-scene-view.
7. Console: console-clear-logs, console-get-logs.
8. Editor state and selection: editor-application-get-state, editor-application-set-state, editor-selection-get, editor-selection-set.
9. Reflection: reflection-method-call, reflection-method-find (can find C# methods including private methods).
10. Script: script-delete, script-execute, script-read, script-update-or-create.
11. Tests: tests-run (supports EditMode/PlayMode filtering and detailed results).
12. Types: type-get-json-schema (generates a JSON Schema for a C# type via reflection).
13. Profiler: profiler-capture-frame, profiler-clear-data, profiler-enable-module, profiler-get-memory-stats, profiler-get-rendering-stats, profiler-get-script-stats, profiler-get-status, profiler-list-modules, profiler-load-data, profiler-save-data, profiler-start, profiler-stop.

Usage (official three steps) Step one: Install the plugin. Method one: Download the .unitypackage installer and import it into the Unity project (you can double-click the file to have Unity open it automatically, or first open the Unity editor, then select the file via Assets/Import Package/Custom Package). Method two (recommended): Run openupm add com.ivanmurzak.unity.mcp, or use the CLI: npx unity-mcp-cli install-plugin ./MyUnityProject. The official note is that the project path cannot contain spaces. Step two: Choose and install an AI agent, such as Claude Code (recommended), Claude Desktop, GitHub Copilot in VS Code, Cursor, or another supported client. The official instructions state that you do not need to install all of them; just choose one as the main chat window for communicating with the LLM. The key requirement is that the AI agent must support Skills or dynamic MCP tool updates. Step three: Configure the AI agent. Open the Unity project and go to Window/AI Game Developer; method one: Click Auto-generate Skills (recommended); Option two: click Configure Model Context Protocol (MCP). If the MCP client you are using is not in the list, you can use the raw JSON displayed in the window to manually inject it into that client. You can also use the command line: npx unity-mcp-cli setup-skills claude-code ./MyUnityProject. After completion, you can make requests to the AI such as "Create 3 cubes in a circle with radius 2".

Manual configuration key points: choose <command> according to your environment; replace <unityProjectPath> with the full path to the Unity project; replace <port> with the port in the AI Game Developer configuration; add the MCP server on the command line, with server name ai-game-developer and type local.

CLI and cloud notes: the official documentation states that installing the plugin via unity-mcp-cli does not require the Unity Editor. unity-mcp-cli login performs a browser OAuth device flow and stores credentials in shared machine storage (~/.ai-game-dev/credentials.json), with no need to paste a personal access token. By default, the AI agent configuration written by setup-mcp points to the project-pinned endpoint https://ai-game.dev/mcp/p/<pin>; pass --no-pin to use the shared endpoint https://ai-game.dev/mcp。团队可通过注册码分发访问权限：unity-mcp-cli install-plugin --enroll <code>.

Team update notification settings: when a new version is available on OpenUPM, the plugin shows an update popup at editor startup, and by default each team member must click "Do not show again" individually. For multi-person Unity projects, the engineer responsible for plugin version management can open Edit ▸ Project Settings ▸ AI Game Developer and enable "Disable update notifications for the entire team"; this setting is saved to ProjectSettings/AI-Game-Developer-UpdateSettings.asset. It only needs to be set once per project; after committing that file, members who pull that commit will suppress the popup. The same toggle can also be accessed via Tools ▸ AI Game Developer ▸ Updates ▸ Disable Update Notifications (Team).

Server deployment: the documentation includes sections on Unity MCP Server setup, plugin variables, Docker, streamableHttp transport, stdio transport, custom ports, and binary executables.

Architecture notes: the documentation explains what MCP is, AI agents, MCP Server, MCP Tool (and when to use it), MCP Resource (and when to use it), and MCP Prompt (and when to use it).

Extension and customization: you can add custom Tools and custom MCP Prompts; when you need more tools, you can install extensions or create your own tools.

Applicable scenarios: automated editor operations, code generation, testing, debugging, and performance profiling in Unity game development, as well as using an LLM in the compiled game to implement dynamic NPC behavior or player-AI interaction.

Limitations and notes: the official documentation requires that the project path must not contain spaces; it requires the AI agent to support Skills or dynamic MCP tool updates; the official documentation states that it collaborates for free with Claude Code, Gemini, Copilot, Cursor, etc., but does not provide specific charging, pricing, or platform restriction information, and none is currently available; other feature details and specific version requirements are subject to the official documentation.