unity
Enables AI assistants to control the Unity Editor via MCP, allowing scene manipulation, asset management, compilation, and testing through natural language.
README
LLM Dev Tools
Unity Editor tooling for AI-assisted development. AgentBridge exposes a
file-based command protocol so external tools (Claude Code, scripts, CI) can
drive the Unity Editor — and MCP turns every command into a tool your LLM can
call directly.
Quickstart
1. Install the package
Window → Package Manager → + → Add package from git URL
https://github.com/simonwittber/AgentBridge.git
Or for a specific version:
https://github.com/simonwittber/AgentBridge.git#v0.1.0
2. Build the CLI
Download a pre-built binary from the latest release
and place it on your PATH. Or build from source:
cd path/to/AgentBridge/Harness~/dffrnt-agent
go build -o dffrnt-agent . # macOS / Linux
go build -o dffrnt-agent.exe . # Windows
3. Wire into Claude Code
Add to .claude/settings.json inside your Unity project:
{
"mcpServers": {
"unity": {
"command": "dffrnt-agent",
"args": ["serve"]
}
}
}
Run dffrnt-agent from your Unity project root, or pass --project <path> to
point it at the project directory.
4. Open Unity and verify
Start (or focus) your Unity project. In Claude Code, the unity MCP server
will appear in /mcp with all bridge commands available as tools.
5. Send your first command
dffrnt-agent status
Expected output:
{
"cmd": "status",
"status": "ok",
"uptime_s": 42.3,
"queued": 0,
"busy": false
}
Requirements
- Unity 6000.0 or later
- Go 1.25+ (only needed to build from source)
Built-in commands
Core
| Command | Description |
|---|---|
status |
Bridge liveness, uptime, queue depth |
compile |
Request script compilation; returns structured errors and warnings |
refresh |
Trigger AssetDatabase.Refresh() and wait for completion |
commands |
List all available commands and their arguments |
Scene
| Command | Description |
|---|---|
scene_info |
Name, path, dirty flag, root count |
scene_open |
Open a scene by asset path |
scene_save |
Save the active scene |
scene_new |
Create a new empty or default scene |
Hierarchy & objects
| Command | Description |
|---|---|
hierarchy |
Scene tree as JSON (configurable depth) |
object_find |
Find a GameObject by path; returns components |
objects_find |
Find all objects with a given component type |
object_create |
Create a GameObject or primitive |
object_delete |
Delete a GameObject |
object_active |
Activate or deactivate a GameObject |
object_rename |
Rename a GameObject |
object_select |
Select one or more objects in the editor |
Components & assets
| Command | Description |
|---|---|
component_get |
Get all serialized fields of a component |
component_set |
Set a serialized field on a component |
component_add |
Add a component by type name |
prefab_open |
Open a prefab in prefab stage |
prefab_save |
Save and exit the current prefab stage |
asset_info |
GUID and importer settings for an asset |
asset_set |
Set an importer field and reimport |
asset_find |
Find assets by type / label filter |
material_get |
Get all shader properties of a material |
material_set |
Set a shader property on a material |
Editor & console
| Command | Description |
|---|---|
console_logs |
Recent Unity console messages (ring buffer, newest first) |
play_mode |
Enter, exit, or query play mode |
menu_item |
Invoke a Unity menu item by path |
run_tests |
Run edit-mode or play-mode tests; returns pass/fail/skip |
uuid |
Generate a UUID v4 |
Adding custom commands
Implement IAgentCommand in any Editor assembly:
using LLMDevTools;
using UnityEditor;
[InitializeOnLoad]
public class MyCommand : IAgentCommand
{
static MyCommand() => AgentBridge.Register(new MyCommand());
public string Cmd => "my_cmd";
public string Description => "Does something useful.";
public ArgSpec[] Args => new[]
{
new ArgSpec("message", "string", "", "Text to log"),
};
public void Execute(string uid, string requestJson)
{
AgentBridge.NewWriter(uid, Cmd)
.Set("echoed", requestJson)
.Send("ok");
}
}
Protocol
Commands are newline-delimited JSON written to Temp/agent_input:
{"uid":"a1b2c3d4","cmd":"compile"}
Responses are appended to Temp/agent_output:
{"uid":"a1b2c3d4","cmd":"compile","status":"ok","session_id":1749123456789,"errors":[],"warnings":[]}
Unity also writes Temp/agent_session every 5 seconds:
{"pid":12345,"state":"idle","active_scene":"Main","play_mode":false,"compile_errors":0,"written_at":1749123456789}
dffrnt-agent reads this file to verify Unity is alive before sending any command.
Output rotates at 2 MB; input is truncated on Unity startup.
LLM Agent Log window
Open via Window → General → LLM Agent Log. Live scrolling view of all commands and responses — green = ok, red = error.
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。