korone-mcp
An MCP bridge that lets AI assistants control Roblox Studio 2021 in real time, offering tools for workspace inspection, instance creation, and Lua execution.
README
<div align="center">
<img src="https://files.catbox.moe/ztl230.png" alt="korone studio 0.477 (2021M) MCP" width="96" height="96">
Korone Studio (2021M) MCP
An MCP bridge that lets AI assistants control Roblox Studio 2021 in real time
</div>
Features
- MCP protocol — speaks the Model Context Protocol, so any MCP-compatible AI assistant can talk to Studio directly.
- Web control panel — a built-in dashboard at
http://127.0.0.1:4444to run tools, browse results and manage connected Studio clients. - 30+ tools — from
read_workspaceandcreate_instancetoexecute_lua,set_lightingand tagging. - Dot-path notation — address instances as
Workspace.Bridge.Partinstead of fighting slash paths. - Multiple clients — several Studio windows can connect at once; each gets its own toggle, and you can target a specific one (click its chip in the panel, or pass
clientto any tool from an MCP client) - Hot reload — while developing, edit
korone-mcp.luaand the plugin picks the changes up in seconds.
Architecture
graph LR
A[AI assistant / MCP client] -->|MCP| B[index.js server]
B -->|HTTP :4444| C[korone-mcp.lua in Studio]
C -->|poll / result| D[Roblox Studio ~2021]
E[Web panel :4444] --> B
Requirements
- Node.js 18+
- Roblox Studio 2021 — any 2021M-era build
- The plugin file
korone-mcp.lualoaded into Studio
Installation
-
Install dependencies and start the server:
npm install npm start -
Open
http://127.0.0.1:4444in a browser to reach the web panel. -
In Studio, load
korone-mcp.luaas a plugin (Plugins folder or drag-and-drop into Studio). -
The panel should switch to connected and the small
MCP: checking...widget appears in the corner.
Configuration
Settings are stored per-user via plugin:SetSetting and can be changed from the plugin panel.
| Setting | Default | Description |
|---|---|---|
MCP_ServerURL |
http://127.0.0.1:4444 |
Base URL of the bridge server |
MCP_ClientName |
studio |
Client key shown on the web panel |
MCP_Enabled |
true |
Master switch for the plugin |
MCP_PollInterval |
0.3 |
Seconds between HTTP polls |
MCP_ShowStatusWidget |
true |
Show/hide the small MCP: checking... widget |
MCP_StatusCheckInterval |
2 |
Seconds between status pings to the panel |
MCP_VerboseLogging |
false |
Extra logging in the Studio output |
Path notation
Instances are addressed with dot notation:
Workspace.Bridge.Part
Workspace.Models.Car.Body
Slash notation (Workspace/Bridge/Part) is also accepted for compatibility.
Studio compatibility
Designed for 2021-era Roblox Studio builds (like 2021M). Things to keep in mind:
- Old
HttpService— only two-argument calls are safe:GetAsync(url)andPostAsync(url, body). Extra arguments (headers, content type) are not supported and fail silently. - ASCII-only strings — non-ASCII characters in the Lua source (like
—or═) turn into mojibake (—) in logs. The plugin keeps its strings ASCII.
HTTP endpoints
| Endpoint | Method | Purpose |
|---|---|---|
/hello |
POST | Plugin registers itself with the server |
/poll |
GET | Plugin asks for pending tool calls |
/result |
POST | Plugin returns a tool result |
/status |
GET | Health/status of the server |
/log |
POST | Plugin forwards Studio log lines |
/call |
POST | Run a tool from the web panel |
/client-toggle |
POST | Enable/disable a specific client by key |
/plugin-update |
GET | Hot-reload source for the plugin |
Usage
- Start the server (
npm start) and load the plugin into Studio. - Open the web panel at
http://127.0.0.1:4444. - Pick a tool, fill in its arguments (placeholders show examples), hit Run →.
- The result appears in the panel; Copy result copies it to your clipboard.
Example: ask your MCP client to run create_instance with parent=Workspace and className=Part, then check read_workspace to see it.
With several Studio windows connected: click a client chip in the panel to target it (highlighted green), or pass client: "studio" to any tool from an MCP client. Without a target, the command goes to the only active client — or, if several are active, to the most recently active one.
Security
By default the bridge is open — it only listens on 127.0.0.1, so only local processes can reach it. If you want extra protection (e.g. a page in your browser could trigger /call while you have the panel open), flip the flag in index.js:
const ENABLE_AUTH = true; // protect /call and /client-toggle
const MCP_TOKEN = 'your-secret';
When enabled, /call and /client-toggle require ?token=... or the x-mcp-token header. Open the panel as http://127.0.0.1:4444/?token=your-secret (the panel stores it and sends it with every request). Keep ENABLE_AUTH = false to disable the check entirely.
Troubleshooting
| Symptom | Fix |
|---|---|
| Panel says checking... forever | Make sure the server is running and MCP_ServerURL points at http://127.0.0.1:4444; check the Studio output for errors |
— / mojibake in Studio logs |
Non-ASCII characters in korone-mcp.lua — replace them with ASCII (-, =) |
| Duplicate/ghost clients on the panel | Restart the server so stale registrations clear; each Studio window uses its own ?client= key |
License
MIT </invoke>
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。