Unity MCP Search
Enables AI assistants to search and analyze Unity project assets, including references and dependencies, via the Model Context Protocol.
README
Unity MCP Search
A Unity Editor package that exposes asset search, references, and dependency analysis as an MCP (Model Context Protocol) server, enabling AI coding assistants like Claude Code and OpenCode to query your Unity project's asset graph.
Architecture
AI Client (Claude Code / OpenCode / ...)
| MCP Protocol (stdio)
v
Python MCP Server (server.py)
| HTTP (localhost:8090)
v
Unity Editor HTTP Server (McpHttpServer.cs)
|
v
Unity AssetDatabase & SearchService APIs
The Python process acts as a thin MCP-to-HTTP bridge. The real work happens inside the Unity Editor on the main thread, using AssetDatabase and SearchService APIs.
Requirements
- Unity 2021.3 or later
- Python 3.10 or later
Installation
Option A: Unity Package Manager (Git URL)
- Open Unity, go to Window > Package Manager
- Click + > Add package from git URL...
- Enter:
https://github.com/StromKuo/Unity-MCP-Search.git
Option B: Git Submodule
git submodule add https://github.com/StromKuo/Unity-MCP-Search.git Packages/com.strodio.unity-mcp-search
Option C: Local Clone
Clone the repo into your project's Packages/ folder:
cd YourProject/Packages
git clone https://github.com/StromKuo/Unity-MCP-Search.git com.strodio.unity-mcp-search
Setup
1. Setup Python Environment
Go to Tools > MCP Search > Setup Python Environment.
This will:
- Find a suitable Python 3.10+ interpreter on your system
- Create a virtual environment inside the package (
MCP~/venv/) - Install the Python dependencies (
mcp,httpx)
2. Configure Your AI Client
Go to Tools > MCP Search > Copy MCP Config to copy the MCP server configuration JSON to your clipboard.
The config looks like this:
{
"mcpServers": {
"unity-search": {
"command": "/path/to/Packages/com.strodio.unity-mcp-search/MCP~/venv/bin/python",
"args": ["/path/to/Packages/com.strodio.unity-mcp-search/MCP~/server.py"]
}
}
}
Paste it into your AI client's MCP settings:
- Claude Code:
~/.claude/settings.json - OpenCode:
~/.config/opencode/config.json(under themcp_serverssection)
3. Verify
Go to Tools > MCP Search > Check Environment to verify everything is set up correctly. You should see:
System Python 3.10+: OK
Virtual Env: OK
Dependencies: OK
HTTP Server: Running (port 8090)
Available MCP Tools
Once configured, your AI client can use the following tools:
search_assets
Search for assets using Unity Search query syntax.
search_assets(query="t:Material sky")
Supported filters:
t:Type— filter by asset type (e.g.t:Texture,t:Prefab,t:Scene)l:label— filter by asset labelref:path— find assets referencing a given pathdep:path— find assets that depend on a given path
Returns: path, name, type for each match.
get_asset_info
Get detailed info about a specific asset.
get_asset_info(asset_path="Assets/Materials/Default.mat")
Returns: path, guid, type, size, sizeFormatted, dependencyCount, referenceCount, dependencies, references.
find_asset_references
Find all assets that reference the specified asset.
find_asset_references(asset_path="Assets/Sprites/hero.png")
Returns a list of asset paths.
find_asset_dependencies
Find all dependencies of the specified asset.
find_asset_dependencies(asset_path="Assets/Prefabs/Enemy.prefab", recursive=True)
Returns a list of asset paths.
find_unused_assets
Scan a directory for assets that are not referenced by anything in the project.
find_unused_assets(directory="Assets/Art", extensions="png,jpg,mat")
Automatically excludes:
Resources/folders (loaded at runtime by name)- Build scenes
- Addressable asset entries
StreamingAssets/- Scripts and shaders
Editor/foldersPackages/
Returns: path, size, sizeFormatted for each unused asset.
Unity Editor Menu
All menu items are under Tools > MCP Search:
| Menu Item | Description |
|---|---|
| Start Server | Start the HTTP server (auto-starts on editor launch) |
| Stop Server | Stop the HTTP server |
| Setup Python Environment | Create venv and install dependencies |
| Check Environment | Verify all components are working |
| Copy MCP Config | Copy MCP server config JSON to clipboard |
| Server Status | Show current server status |
How It Works
- The Unity HTTP server (
McpHttpServer.cs) starts automatically when the editor opens via[InitializeOnLoad]. It listens onlocalhost:8090. - HTTP requests from the Python bridge are queued and processed on Unity's main thread (required by
AssetDatabaseandSearchServiceAPIs). - The Python MCP server (
MCP~/server.py) translates MCP tool calls into HTTP requests. TheMCP~directory is ignored by Unity's asset importer (directories ending with~are excluded). - The Python venv lives inside the package at
MCP~/venv/and is excluded from version control via.gitignore.
Troubleshooting
"Cannot connect to Unity Editor"
- Make sure Unity Editor is open and focused (the HTTP server runs in the editor process)
- Check Tools > MCP Search > Server Status
- Try Tools > MCP Search > Stop Server, then Start Server
"Asset not found" errors
- Asset paths must use forward slashes and start with
Assets/(e.g.Assets/Sprites/hero.png)
Python setup fails
- Ensure Python 3.10+ is installed:
python3 --version - On macOS with Homebrew:
brew install python@3.12 - On Windows: download from https://www.python.org/downloads/
Timeouts on large projects
find_unused_assetson the rootAssets/directory may take a while on large projects. Use thedirectoryparameter to narrow the scope, orextensionsto filter by file type.
License
MIT
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。
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。
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。
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。