Unity Strands Agent

基于 AWS Strands Agents SDK 的 Unity 编辑器 AI 插件，在 Unity 编辑器中直接与 AI 对话，获得问题解答、学习指导和开发建议。

Language / 语言：中文 | English

中文版本

Unity Strands Agent 是一个友好易用的 AI 助手，将 AWS 开源的 Strands Agents SDK 无缝集成到 Unity 编辑器中。该插件内置了对 Unity 项目的深度理解，能够通过自然语言对话帮助您学习 Unity、解决问题、理解项目结构。通过支持 Model Context Protocol (MCP)，本插件可以与 mcp-unity 或其他 Unity MCP 插件灵活搭配，让您轻松上手 Unity 开发，大幅降低学习门槛。

核心实现

基于 Strands Agents SDK 开发智能助手极其简单，核心代码展示了其优雅的设计：

from strands import Agent
from unity_tools import get_unity_tools

class UnityAgent:
    def __init__(self):
        # 初始化 MCP 管理器
        from mcp_manager import MCPManager
        self.mcp_manager = MCPManager()
        
        # 从配置文件加载 MCP 服务器并建立连接
        # 配置示例 (Assets/UnityAIAgent/mcp_config.json):
        # {
        #   "mcpServers": {
        #     "mcp-unity": {
        #       "command": "/usr/local/bin/node",
        #       "args": ["path/to/mcp-unity/Server/build/index.js"],
        #       "env": {"UNITY_PORT": "8090"}
        #     }
        #   }
        # }
        mcp_tools = self.mcp_manager.load_mcp_tools()
        
        # 获取 Unity 开发工具集（包含内置工具和 MCP 工具）
        unity_tools = get_unity_tools(include_mcp=True, agent_instance=self)
        
        # 创建强大的 AI Agent，配备完整工具集
        self.agent = Agent(
            system_prompt=UNITY_SYSTEM_PROMPT, 
            tools=unity_tools  # 21+内置工具 + 动态加载的 MCP 工具
        )
    
    def _load_mcp_tools(self):
        # 由 get_unity_tools 调用，返回已加载的 MCP 工具
        return self.mcp_manager._mcp_tools
    
    def process_message(self, message: str):
        # 处理用户消息，支持 Unity 原生操作
        return self.agent(message)
    
    def cleanup(self):
        # 清理 MCP 连接和资源
        self.mcp_manager.cleanup()

通过以上代码可以看到：

MCP 配置加载：从 JSON 配置文件读取 MCP 服务器设置
动态工具发现：MCP 服务器启动后自动发现并加载可用工具
工具集整合：将内置工具与 MCP 工具无缝整合
资源管理：提供清理机制确保连接正确关闭

该插件基于模块化架构设计，通过 MCP 协议轻松扩展工具集，让 AI Agent 能够直接操作 Unity 编辑器。

主要特性

Unity 专业化

项目结构理解：智能识别 Unity 项目结构，提供清晰易懂的解释和指导
学习助手：耐心解答 Unity 概念、工作流程和最佳实践，适合各种经验水平
问题解决：快速诊断常见问题，提供详细的解决步骤和方案
编辑器操作：通过 MCP 协议与 Unity MCP 插件（如 mcp-unity）搭配，实现场景操作、资源管理等直观交互

简单易用

自然语言交互：用日常语言描述您的需求，无需学习复杂的技术术语
一键安装配置：自动检测环境、安装依赖，几分钟即可开始使用
智能引导：根据您的问题提供循序渐进的学习建议和操作指导
MCP 协议支持：通过 Model Context Protocol 灵活扩展能力，可与 mcp-unity 或其他 Unity MCP 插件无缝集成

智能助手

24/7 在线支持：随时为您解答 Unity 相关问题，从基础概念到高级技巧
个性化学习：根据您的项目和需求，提供定制化的学习路径和建议
持续改进：通过与您的对话不断学习，提供越来越精准的帮助

AI 模型支持

多模型兼容：Strands Agents SDK 支持 Claude、GPT、Llama 等主流 AI 模型
当前支持：本插件当前版本仅支持 Bedrock US Oregon (us-west-2) 区域的 Claude 3.7 Sonnet
智能记忆：记住您的学习进度和项目偏好，提供个性化学习建议

系统架构

%%{init: {'theme':'neutral'}}%%
graph TD
    subgraph Unity["Unity 编辑器"]
        UI[Unity AI Agent Window<br/>用户界面]
        PB[Python Bridge<br/>Python桥接层]
    end
    
    subgraph AgentPython["智能 Agent 系统"]
        AC[Agent Core<br/>Agent 核心]
        UA[Unity Agent<br/>Unity 专用 Agent]
        SDK[Strands Agents SDK<br/>核心AI框架]
    end
    
    subgraph ModelServices["AI 模型服务"]
        AWS[Amazon Bedrock<br/>✓ 插件已支持]
        ANTHROPIC[Anthropic Claude<br/>SDK支持]
        OPENAI[OpenAI GPT<br/>SDK支持]
        OLLAMA[Ollama 本地模型<br/>SDK支持]
        OTHERS[LiteLLM / Llama 等<br/>SDK支持]
    end
    
    subgraph MCPServices["扩展服务生态"]
        MCP[MCP Servers<br/>扩展服务]
    end
    
    %% Connections
    UI -->|用户交互| PB
    PB -->|Python.NET| AC
    AC -->|创建| UA
    UA ==>|依赖| SDK
    SDK ==>|调用| AWS
    SDK -.->|支持但未集成| ANTHROPIC
    SDK -.->|支持但未集成| OPENAI
    SDK -.->|支持但未集成| OLLAMA
    SDK -.->|支持但未集成| OTHERS
    SDK -->|调用| MCP
    
    %% Subgraph styles
    style Unity fill:#E3F2FD,stroke:#1976D2,stroke-width:2px
    style AgentPython fill:#E3F2FD,stroke:#1976D2,stroke-width:2px
    style ModelServices fill:#E3F2FD,stroke:#1976D2,stroke-width:2px
    style MCPServices fill:#E3F2FD,stroke:#1976D2,stroke-width:2px
    
    %% Node styles - Blue for supported features
    style UI fill:#BBDEFB,stroke:#1976D2,stroke-width:2px
    style PB fill:#BBDEFB,stroke:#1976D2,stroke-width:2px
    style AC fill:#BBDEFB,stroke:#1976D2,stroke-width:2px
    style UA fill:#1976D2,stroke:#0D47A1,stroke-width:3px,color:#FFF
    style AWS fill:#BBDEFB,stroke:#1976D2,stroke-width:2px
    style MCP fill:#BBDEFB,stroke:#1976D2,stroke-width:2px
    
    %% Core SDK - Red for emphasis
    style SDK fill:#D32F2F,stroke:#B71C1C,stroke-width:4px,color:#FFF
    
    %% SDK Only services - Gray
    style ANTHROPIC fill:#EEEEEE,stroke:#9E9E9E,stroke-width:1px
    style OPENAI fill:#EEEEEE,stroke:#9E9E9E,stroke-width:1px
    style OLLAMA fill:#EEEEEE,stroke:#9E9E9E,stroke-width:1px
    style OTHERS fill:#EEEEEE,stroke:#9E9E9E,stroke-width:1px

快速开始

3分钟快速上手

安装插件

Unity Package Manager → Add package from git URL
输入：https://github.com/ddpie/unity-strands-agent.git

自动配置

Window → Unity Strands Agent → Settings → Environment Setup → Start Setup
等待自动安装完成（无需任何手动配置）

开始聊天

点击 `Open AI Assistant`，像聊天一样问问题即可
例如："Unity中的GameObject是什么？"

系统要求：Unity 2022.3+ | Python 3.11 | macOS | AWS credentials

视频教程：Unity Strands Agent 功能演示

详细配置

系统要求

Unity 2022.3 LTS 或更高版本
Python 3.11（插件会自动检测和配置）
AWS credentials 配置（默认使用 Amazon Bedrock 服务，默认模型为 US Oregon (us-west-2) 区域的 Claude 3.7 Sonnet）
仅支持 macOS 系统

重要：

本插件目前仅适配 Python 3.11，暂不支持其他 Python 版本。在 macOS 上可通过 brew install python@3.11 安装。
环境初始化将在点击 Environment Setup 的 Start Setup 按钮时进行，请耐心等待安装过程完成。

安装步骤

通过 Unity Package Manager 安装：

在 Unity 中打开 Package Manager
点击 "+" 按钮
选择 "Add package from git URL"
输入：https://github.com/ddpie/unity-strands-agent.git
点击 "Add" 并等待 Unity 自动下载和导入

配置步骤

在 Unity 中选择菜单：Window → Unity Strands Agent
点击 Settings 标签页，然后点击 Environment Setup
点击 Start Setup 按钮开始自动安装
系统将自动完成环境检测、虚拟环境创建、依赖安装等步骤
配置完成后，点击 Open AI Assistant 开始使用

环境配置

插件会自动设置以下环境变量：

PROJECT_ROOT_PATH：项目根目录路径
STRANDS_TOOLS_PATH：Strands工具路径

配置文件保存在 Assets/UnityAIAgent/PathConfiguration.asset 中，会自动加载。插件使用 AWS credentials 配置文件访问 Bedrock 服务。

MCP 服务器配置：本插件支持通过 MCP 协议与各种 Unity MCP 插件搭配使用。在 Unity 编辑器界面的 Settings → MCP Configuration 中以 JSON 格式添加服务器配置。建议使用绝对路径配置 command 字段，例如配置 mcp-unity 插件：

{
  "mcpServers": {
    "mcp-unity": {
      "command": "/usr/local/bin/node",
      "env": {
        "UNITY_PORT": "8090"
      },
      "args": [
        "/path/to/your/unity-project/Library/PackageCache/com.gamelovers.mcp-unity@latest/Server/build/index.js"
      ]
    }
  }
}

您可以根据需求选择和配置不同的 Unity MCP 插件。关于 mcp-unity 插件的详细安装和使用说明，请参考：mcp-unity 插件文档

故障排除

菜单项不显示

检查Unity控制台是否有编译错误
确认包已正确下载到Library/PackageCache目录
重启Unity编辑器

Python 3.11 环境问题

确认系统已安装 Python 3.11：python3.11 --version
在 macOS 上安装：brew install python@3.11
重新运行 Environment Setup 进行自动配置

找不到agent_core模块

运行 Environment Setup 重新安装Python依赖
检查Library/PackageCache中是否有com.ddpie.unity-strands-agent包
确认虚拟环境在项目Python目录中正确创建

如有问题，请检查Unity控制台的错误日志，并在设置页面点击 验证配置 查看当前配置状态。

支持和反馈

文档：Strands Agents SDK 官方文档
问题反馈：GitHub Issues

参考资料

Strands Agents SDK

官方文档：https://strandsagents.com/latest/
Python SDK：https://github.com/strands-agents/sdk-python
示例代码：https://github.com/strands-agents/samples
工具包：https://github.com/strands-agents/tools
Agent Builder：https://github.com/strands-agents/agent-builder

技术实现

Python.NET：https://github.com/pythonnet/pythonnet
Model Context Protocol (MCP)：https://modelcontextprotocol.io/
Unity Editor 扩展：https://docs.unity3d.com/Manual/ExtendingTheEditor.html

English Version

Unity Strands Agent is a friendly and accessible AI assistant that seamlessly integrates AWS's open-source Strands Agents SDK into the Unity Editor. Built with deep understanding of Unity project structures and workflows, this plugin helps you learn Unity, solve problems, and understand project organization through natural language conversations. Through support for the Model Context Protocol (MCP), it can be flexibly paired with mcp-unity or other Unity MCP plugins to make Unity development approachable and significantly lower the learning curve.

Core Implementation

Building intelligent assistants with the Strands Agents SDK is remarkably simple. The core implementation showcases its elegant design:

from strands import Agent
from unity_tools import get_unity_tools

class UnityAgent:
    def __init__(self):
        # Initialize MCP manager
        from mcp_manager import MCPManager
        self.mcp_manager = MCPManager()
        
        # Load MCP servers from config and establish connections
        # Config example (Assets/UnityAIAgent/mcp_config.json):
        # {
        #   "mcpServers": {
        #     "mcp-unity": {
        #       "command": "/usr/local/bin/node",
        #       "args": ["path/to/mcp-unity/Server/build/index.js"],
        #       "env": {"UNITY_PORT": "8090"}
        #     }
        #   }
        # }
        mcp_tools = self.mcp_manager.load_mcp_tools()
        
        # Load Unity development toolset (includes built-in and MCP tools)
        unity_tools = get_unity_tools(include_mcp=True, agent_instance=self)
        
        # Create a powerful AI Agent with complete toolset
        self.agent = Agent(
            system_prompt=UNITY_SYSTEM_PROMPT, 
            tools=unity_tools  # 21+ built-in tools + dynamically loaded MCP tools
        )
    
    def _load_mcp_tools(self):
        # Called by get_unity_tools to retrieve loaded MCP tools
        return self.mcp_manager._mcp_tools
    
    def process_message(self, message: str):
        # Process user messages with Unity native operations support
        return self.agent(message)
    
    def cleanup(self):
        # Cleanup MCP connections and resources
        self.mcp_manager.cleanup()

The code above demonstrates:

MCP Configuration Loading: Reads MCP server settings from JSON configuration files
Dynamic Tool Discovery: Automatically discovers and loads available tools after MCP server startup
Tool Integration: Seamlessly combines built-in tools with MCP tools
Resource Management: Provides cleanup mechanisms to ensure connections are properly closed

Built on a modular architecture, this plugin leverages the MCP protocol to extend the toolset, enabling the AI Agent to directly operate the Unity Editor.

Key Features

Unity Specialization

Project Understanding: Intelligently analyzes your Unity project structure and provides clear, easy-to-understand explanations
Learning Companion: Patiently explains Unity concepts, workflows, and best practices suitable for all experience levels
Problem Solver: Quickly diagnoses common issues and provides detailed step-by-step solutions
Editor Operations: Through MCP protocol integration with Unity MCP plugins (such as mcp-unity), enables intuitive scene operations, asset management, and more

Simple and Accessible

Natural Language Interaction: Describe your needs in everyday language without learning complex technical terminology
One-Click Setup: Automatically detects environment, installs dependencies, ready to use in minutes
Intelligent Guidance: Provides step-by-step learning suggestions and operation guidance based on your questions
MCP Protocol Support: Flexibly extend capabilities through the Model Context Protocol, seamlessly integrating with mcp-unity or other Unity MCP plugins

Intelligent Assistant

24/7 Support: Always available to answer your Unity questions, from basic concepts to advanced techniques
Personalized Learning: Provides customized learning paths and suggestions based on your projects and needs
Continuous Improvement: Learns from your conversations to provide increasingly accurate and helpful assistance

AI Model Support

Multi-Model Compatibility: The Strands Agents SDK supports Claude, GPT, Llama, and other leading AI models
Current Support: This plugin currently only supports Claude 3.7 Sonnet in Bedrock US Oregon (us-west-2) region
Intelligent Memory: Automatically remembers your project patterns, coding style, and preferences across sessions

System Architecture

%%{init: {'theme':'neutral'}}%%
graph TD
    subgraph Unity["Unity Editor"]
        UI[Unity AI Agent Window<br/>User Interface]
        PB[Python Bridge<br/>Integration Bridge]
    end
    
    subgraph AgentPython["AI Agent System"]
        AC[Agent Core<br/>Core Agent System]
        UA[Unity Agent<br/>Unity Specialized Agent]
        SDK[Strands Agents SDK<br/>Core AI Framework]
    end
    
    subgraph ModelServices["AI Model Services"]
        AWS[Amazon Bedrock<br/>✓ Plugin Ready]
        ANTHROPIC[Anthropic Claude<br/>SDK Only]
        OPENAI[OpenAI GPT<br/>SDK Only]
        OLLAMA[Ollama Local Models<br/>SDK Only]
        OTHERS[LiteLLM / Llama etc<br/>SDK Only]
    end
    
    subgraph MCPServices["Extension Ecosystem"]
        MCP[MCP Servers<br/>Extension Services]
    end
    
    %% Connections
    UI -->|User Interaction| PB
    PB -->|Python.NET| AC
    AC -->|Creates| UA
    UA ==>|Depends on| SDK
    SDK ==>|Calls| AWS
    SDK -.->|Supports but not integrated| ANTHROPIC
    SDK -.->|Supports but not integrated| OPENAI
    SDK -.->|Supports but not integrated| OLLAMA
    SDK -.->|Supports but not integrated| OTHERS
    SDK -->|Calls| MCP
    
    %% Subgraph styles
    style Unity fill:#E3F2FD,stroke:#1976D2,stroke-width:2px
    style AgentPython fill:#E3F2FD,stroke:#1976D2,stroke-width:2px
    style ModelServices fill:#E3F2FD,stroke:#1976D2,stroke-width:2px
    style MCPServices fill:#E3F2FD,stroke:#1976D2,stroke-width:2px
    
    %% Node styles - Blue for supported features
    style UI fill:#BBDEFB,stroke:#1976D2,stroke-width:2px
    style PB fill:#BBDEFB,stroke:#1976D2,stroke-width:2px
    style AC fill:#BBDEFB,stroke:#1976D2,stroke-width:2px
    style UA fill:#1976D2,stroke:#0D47A1,stroke-width:3px,color:#FFF
    style AWS fill:#BBDEFB,stroke:#1976D2,stroke-width:2px
    style MCP fill:#BBDEFB,stroke:#1976D2,stroke-width:2px
    
    %% Core SDK - Red for emphasis
    style SDK fill:#D32F2F,stroke:#B71C1C,stroke-width:4px,color:#FFF
    
    %% SDK Only services - Gray
    style ANTHROPIC fill:#EEEEEE,stroke:#9E9E9E,stroke-width:1px
    style OPENAI fill:#EEEEEE,stroke:#9E9E9E,stroke-width:1px
    style OLLAMA fill:#EEEEEE,stroke:#9E9E9E,stroke-width:1px
    style OTHERS fill:#EEEEEE,stroke:#9E9E9E,stroke-width:1px

Quick Start

3-Minute Quick Start

Install Plugin

Unity Package Manager → Add package from git URL
Enter: https://github.com/ddpie/unity-strands-agent.git

Auto Setup

Window → Unity Strands Agent → Settings → Environment Setup → Start Setup
Wait for automatic installation (no manual configuration needed)

Start Chatting

Click `Open AI Assistant` and ask questions like chatting
Example: "What is a GameObject in Unity?"

Requirements: Unity 2022.3+ | Python 3.11 | macOS | AWS credentials

Video Demo: Unity Strands Agent Demo

Detailed Configuration

System Requirements

Unity 2022.3 LTS or higher
Python 3.11 (plugin automatically detects and configures)
AWS credentials configuration (default model provider is Amazon Bedrock with Claude 3.7 Sonnet in the US Oregon (us-west-2) region)
macOS only

Important:

This plugin currently only supports Python 3.11 and does not support other Python versions. On macOS, install with brew install python@3.11.
Environment initialization will occur when you click the Start Setup button in Environment Setup, please wait for the installation process to complete.

Installation Steps

Through Unity Package Manager:

Open Package Manager in Unity
Click "+" button
Select "Add package from git URL"
Enter: https://github.com/ddpie/unity-strands-agent.git
Click "Add" and wait for Unity to automatically download and import

Configuration Steps

In Unity, select menu: Window → Unity Strands Agent
Click Settings tab, then click Environment Setup
Click Start Setup button to begin automatic installation
The system will automatically complete environment detection, virtual environment creation, dependency installation
After configuration completes, click Open AI Assistant to start using

Main capabilities include:

Clear explanations of Unity concepts and project structure
Step-by-step guidance for common Unity tasks and workflows
Helpful suggestions for learning Unity development
Friendly assistance with Unity-specific questions and challenges

Environment Configuration

The plugin automatically sets the following environment variables:

PROJECT_ROOT_PATH: Project root directory path
STRANDS_TOOLS_PATH: Strands tools path

Configuration is saved in Assets/UnityAIAgent/PathConfiguration.asset and will be automatically loaded. The plugin uses AWS credentials configuration files to access Bedrock services.

MCP Server Configuration: This plugin supports pairing with various Unity MCP plugins through the MCP protocol. Configure MCP servers through the Unity editor interface at Settings → MCP Configuration in JSON format. It's recommended to use absolute paths for the command field, for example configuring the mcp-unity plugin:

{
  "mcpServers": {
    "mcp-unity": {
      "command": "/usr/local/bin/node",
      "env": {
        "UNITY_PORT": "8090"
      },
      "args": [
        "/path/to/your/unity-project/Library/PackageCache/com.gamelovers.mcp-unity@latest/Server/build/index.js"
      ]
    }
  }
}

You can choose and configure different Unity MCP plugins based on your needs. For detailed installation and usage instructions for the mcp-unity plugin, please refer to: mcp-unity Plugin Documentation

Troubleshooting

Menu Item Not Showing

Check Unity console for compilation errors
Confirm package has been properly downloaded to Library/PackageCache directory
Restart Unity Editor

Python 3.11 Environment Issues

Confirm Python 3.11 is installed on system: python3.11 --version
Install on macOS: brew install python@3.11
Re-run Environment Setup for automatic configuration

Cannot Find agent_core Module

Run Environment Setup to reinstall Python dependencies
Check if com.ddpie.unity-strands-agent package exists in Library/PackageCache
Confirm virtual environment is correctly created in project Python directory

If you encounter issues, check Unity console for error logs and click Validate Configuration in settings page to view current configuration status.

Support and Feedback

Documentation: Strands Agents SDK Official Documentation
Issue Reporting: GitHub Issues

References

Name		Name	Last commit message	Last commit date
Latest commit History 128 Commits
Assets		Assets
Editor		Editor
Python		Python
.gitignore		.gitignore
Assets.meta		Assets.meta
Editor.meta		Editor.meta
LICENSE		LICENSE
Python.meta		Python.meta
README.md		README.md
package.json		package.json
references.md		references.md
requirement.md		requirement.md

License

ddpie/unity-strands-agent

Folders and files

Latest commit

History

Repository files navigation

Unity Strands Agent

中文版本

核心实现

主要特性

Unity 专业化

简单易用

智能助手

AI 模型支持

系统架构

快速开始

3分钟快速上手

详细配置

系统要求

安装步骤

配置步骤

环境配置

故障排除

支持和反馈

参考资料

Strands Agents SDK

技术实现

相关项目

English Version

Core Implementation

Key Features

Unity Specialization

Simple and Accessible

Intelligent Assistant

AI Model Support

System Architecture

Quick Start

3-Minute Quick Start

Detailed Configuration

System Requirements

Installation Steps

Configuration Steps

Environment Configuration

Troubleshooting

Support and Feedback

References

Strands Agents SDK

Technical Implementation

Related Projects

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Languages