Mengine MCP
Enables AI agents to launch, inspect, control, debug, and hot-reload live Mengine game builds through Visual Studio Code or standard MCP clients.
README
Mengine MCP
Give your AI agent a live connection to your game. Mengine MCP lets Visual Studio Code launch, inspect, control, debug, and hot-reload Development builds made with the open-source Mengine game engine.
Instead of reasoning from source alone, an agent can click UI, inject touch and keyboard input, inspect the active scene, capture frames, read logs, evaluate Python, stop on script breakpoints, and verify changes inside the running game.
Highlights
- Launch existing desktop and mobile builds in visible, hidden-render, or headless-logic mode.
- Inspect and edit the live scene using generation-safe node handles.
- Inject deterministic mouse, keyboard, and multi-touch sequences without moving the system cursor.
- Capture frames, read logs and diagnostics, and wait for runtime conditions.
- Inspect, evaluate, and mutate Python state, including paused debugger frames.
- Explicitly hot-reload Python modules and resource overlays with rollback on failure.
- Extend the public tool surface from game-side C++ or Python handlers.
Mengine MCP does not build the game. It launches artifacts produced by your normal Mengine, IDE, or CI build workflow.
Install in Visual Studio Code
Install Mengine MCP from the Visual Studio Marketplace, or install a locally built VSIX:
code --install-extension mengine-mcp-0.1.0.vsix
The VSIX contains both the extension and the STDIO MCP server. Users do not need to install a separate Node.js runtime.
Configure a game
- Build the game in Development mode with
MCPPluginenabled. The plugin must remain excluded from Master and Release builds. - Open the game root in VS Code.
- Run
Mengine MCP: Create Configurationfrom the Command Palette. - Edit the generated
mengine.mcp.jsonwith the commands and paths for the game. - Open Chat in agent mode. VS Code automatically discovers
Mengine MCP: <workspace>.
The descriptor is resolved relative to the game root. A minimal desktop profile looks like this:
{
"version": 1,
"apps": [
{
"id": "game",
"name": "My Mengine Game",
"profiles": [
{
"id": "macos-dev",
"platform": "macos",
"command": "./build/Debug/MyGame.app/Contents/MacOS/MyGame"
}
],
"scriptRoots": [
{
"path": "./Scripts",
"logicalPrefix": "Scripts"
}
],
"assetRoots": [
{
"path": "./Resources",
"fileGroup": "game"
}
]
}
]
}
See mengine.mcp.example.json for desktop and Android profiles. Android launch profiles receive the listener endpoint and session token as Intent extras, and can use portForwardCommand for adb reverse.
Run the STDIO server directly
The server also works with Codex and other MCP clients independently of VS Code:
npm ci
npm run build
node dist/mengine-mcp.mjs --config /path/to/game/mengine.mcp.json
It speaks standard MCP over STDIO and connects to the running game through the authenticated MNCP v1 TCP protocol.
Development
npm ci
npm test
npm run package
npm run package creates a platform-independent VSIX. Tagged v* releases are verified and published by the Release GitHub Actions workflow using the VSCE_PAT repository secret.
Safety
This extension intentionally has powerful Development-only capabilities, including application launch commands and unrestricted Python execution. VS Code therefore disables it in untrusted workspaces. Session tokens are generated per launch and must never be logged or committed.
See SECURITY.md for the complete security boundary.
Related project
Mengine is a multi-platform C++ game engine used to build and ship graphical games across desktop and mobile platforms. Mengine MCP is its AI-agent development companion.
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。