Game MCP
Multi-agent game development framework for Claude Code that lets you build games with an AI team including Lead, Engineer, Designer, Artist, QA, and Story Consultant.
README
Game MCP
Multi-agent game development framework for Claude Code. Build games with an AI team — Lead, Engineer, Designer, Artist, QA, and Story Consultant.
Quick Start
npx game-mcp
The CLI will guide you through:
- Game name and code
- Game type — Web (HTML/JS), Unity 2D, or Unity 3D
- API keys — Gemini, PixelLab, GitHub/GitLab (stored encrypted, gitignored)
- Platform targets — PC, iOS, Android, Steam
- Git provider — GitHub, GitLab, or local
Then open the project in Claude Code and type /lead to start.
Game Types
Web Game
- Single HTML file with inline CSS/JS
- Canvas API game loop
- Packageable for Steam via Electron
- No dependencies, pure vanilla JS
Unity 2D
- Unity 6 with URP 2D
- PixelLab MCP for AI pixel art generation
- Aseprite CLI for sprite post-processing
- VContainer DI + UniTask async
Unity 3D
- Unity 6 with URP 3D
- Blender MCP for 3D modeling
- PBR materials and lighting
- VContainer DI + UniTask async
How It Works
Game MCP uses a Bible-first approach: design your game completely in documentation before writing code.
Agent Roles
| Command | Role | What They Do |
|---|---|---|
/lead |
Team Lead | Research, plan, create tasks |
/engineer |
Engineer | Write game code |
/designer |
Designer | Write game design docs |
/artist |
Artist | Generate and process art assets |
/qa |
QA | 5-check verification of all changes |
/story |
Story | Narrative structure consulting |
Task Flow
/lead → analyzes request → creates tasks with priorities
/engineer → claims task → implements → self-reviews → completes
/qa → verifies → 5-check gate → approves or reports bugs
Requirements
- Claude Code (CLI)
- Python 3.11+ + uv
- Git
- Node.js 18+ (for CLI and web games)
Optional (by game type)
| Tool | Web | Unity 2D | Unity 3D |
|---|---|---|---|
| Unity 6 | Required | Required | |
| Blender | Required | ||
| Aseprite | Recommended | ||
| Gemini CLI | Recommended | Recommended | Recommended |
API Keys (optional, enable AI features)
| Key | Purpose | Get it at |
|---|---|---|
GOOGLE_API_KEY |
Gemini research + image gen | Google AI Studio |
PIXELLAB_API_TOKEN |
AI pixel art (2D) | PixelLab |
RODIN_API_KEY |
AI 3D models (3D) | Rodin |
GIT_TOKEN |
GitHub/GitLab integration | GitHub/GitLab settings |
Project Structure (after setup)
my-game/
├── .claude/ # Claude Code config + agent commands
│ ├── CLAUDE.md # Project instructions
│ ├── settings.json # Permissions
│ └── commands/ # /lead, /engineer, /designer, /artist, /qa, /story
├── .mcp.json # MCP server config
├── .env # API keys (gitignored)
├── src/ # Game code (HTML or Unity scripts)
├── docs/bible/ # Game design document (13 sections)
├── workflow/ # Task management state
├── game-mcp.json # Project config
└── pyproject.toml # MCP server dependencies
MCP Server
The framework includes a Python MCP server (game-mcp-team) that provides:
- Task management — create, claim, complete, search tasks with priorities
- Decision log — cross-agent context sharing
- Bible status — documentation completeness tracking
- External tools — Gemini CLI, Aseprite CLI, background removal
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。
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。