ui-toolkit-mcp
Enables AI agents to inspect Unity UI Toolkit screens, examine element structure and styles, and capture screenshots without entering Play Mode or building a Player.
README
UI Toolkit MCP
An MCP server that lets AI agents inspect UI Toolkit screens while the Unity Editor remains open. It can examine the structure and appearance of UXML/USS files and capture screenshots without entering Play Mode or building a Player.
Key features:
- List UXML files and open UI Toolkit windows
- Inspect element trees, text, values, positions, sizes, and resolved styles
- Capture an entire screen or a selected element as a PNG image
- Automatically split long screens into tiles (up to 8192 px per image and 100 MP in total)
- Reimport UXML/USS files
- Compatibility tested with Unity 2022.3 LTS and Unity 6.3
Setup
1. Add the Unity package
In the Unity Editor, open Window > Package Manager, click the + button in the upper-left corner, and select Add package from git URL....
Enter the following URL:
https://github.com/adzukyat/ui-toolkit-mcp.git?path=/unity-package
The preview server starts automatically after Unity finishes script compilation. Keep the Unity Editor open while using it.
2. Register the MCP server
Node.js 20 or later is required. Example configuration for Codex:
[mcp_servers.ui_toolkit_mcp]
command = "npx"
args = [
"-y",
"ui-toolkit-mcp-server@latest",
"mcp"
]
MCP tools
| Tool | Purpose |
|---|---|
ui_list_projects |
List open Unity projects that have the package installed |
ui_select_project |
Select the Unity project to work with |
ui_status |
Check the Unity version and connection status |
ui_list_targets |
List preview configurations, UXML files, and open EditorWindow instances |
ui_inspect |
Inspect the structure, position, size, and styles of UI elements |
ui_screenshot |
Capture the UI as a PNG image |
ui_reload |
Reload UXML/USS changes in Unity |
Start with ui_list_projects to find a project, then select it with ui_select_project. Next, use ui_list_targets to find the screen you want to inspect, followed by ui_inspect or ui_screenshot.
Preview configuration (optional)
Place a .ui-toolkit-mcp-preview.json file in the project root to define stable aliases, additional USS files, themes, and viewports. Copy and adapt the following example:
{
"$schema": "./Packages/me.adzuki.ui-toolkit-mcp.preview-server/Schemas/ui-toolkit-mcp-preview.schema.json",
"schemaVersion": 1,
"previews": [
{
"alias": "settings",
"document": "Assets/UI/Settings.uxml",
"stylesheets": ["Assets/UI/Settings.uss"],
"theme": "editor-dark",
"background": "#383838",
"selector": "#settings-root",
"viewport": { "width": 480, "height": "full" }
}
]
}
Supported themes are editor-dark, editor-light, and runtime. background accepts theme, #RRGGBB, or #RRGGBBAA. A panelSettings asset path can also be specified for the runtime theme.
Screenshots use the selected theme's standard canvas color by default (#383838 for Editor Dark and #C8C8C8 for Editor Light). Pass #00000000 as background when a transparent PNG is required. Runtime previews remain transparent unless the UI itself paints a background.
Development and verification
npm test
npm run typecheck
npm run build
node scripts/mcp-smoke.mjs [project-id]
Unity EditMode tests are located in test-projects/2022.3 and test-projects/6000.3. Pixel-accurate verification requires a graphics device. Tree inspection works with -nographics, but PNG rendering does not, so perform final verification in a normally launched Unity Editor.
Current limitations
- The server primarily targets static UI states. For screens constructed later in C#, open the actual
EditorWindowor prepare the required state in a preview UXML file. - The Editor offscreen renderer calls internal UI Toolkit APIs through reflection.
ui_statusreports a warning for unsupported Unity versions. - Open-window IDs are valid only for the current Unity session.
- Pseudo-state injection such as hover and focus, pointer/keyboard interaction, and fixed animation time are not yet supported.
- Fonts, icons, custom controls, and data bindings are rendered only when they can be resolved in the target project.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。