Roblox-MCP-proxy
Enables cloud agents like Notion AI to connect to a local Roblox Studio MCP server via a persistent SSE bridge with authentication.
README
Roblox Studio MCP to Notion Bridge
This project solves the architectural incompatibility between the official Roblox Studio MCP (Model Context Protocol) and stateless cloud agents (like Notion AI).
⚠️ The Problem
The official Roblox Studio MCP runs locally as a stdio executable (StudioMCP.exe).
Tools like Cursor or local Claude apps connect to it directly via child processes. They keep the connection open forever, which maintains the "Active Studio" state.
However, cloud-based agents like Notion communicate over HTTP/SSE. Notion connects, executes a tool, and drops the connection. When it reconnects, it spins up a new local MCP process that has forgotten which Studio instance is active. Because of this, trying to use supergateway out-of-the-box results in crashes, lost state, and the inability to execute complex tasks like getting the DataModel tree.
🛠️ The Solution
This bridge creates a Persistent Multiplexer.
mcp-server.jslaunches a single, global instance ofStudioMCP.exein the background and keeps it alive forever.- It listens for incoming transient SSE connections from the cloud (Notion).
- It intercepts and filters out unsolicited
notifications/tools/list_changedevents that otherwise cause Notion's MCP client to crash. proxy.jsadds a layer of Bearer Token authentication so you can safely expose the bridge to the public internet via a tunnel.
🚀 Setup Instructions
1. Installation
Clone this repository and install the dependencies:
npm install
2. Configuration
Copy the .env.example file to .env:
cp .env.example .env
Open .env and fill in your details:
PUBLIC_URL: The public HTTPS URL where your tunnel is hosted (e.g.https://roblox.your-domain.com). This is required so the SSE endpoint can tell Notion where to send POST messages.AUTH_TOKEN: Create a secure, random string (e.g.my-super-secret-token-123).
3. Start the Bridge
Run the launcher script:
start-bridge.bat
This will open two terminal windows: one for the MCP Server (Port 8000) and one for the Auth Proxy (Port 8001).
4. Create a Tunnel
You must expose Port 8001 to the internet. You can use Cloudflare Tunnels, ngrok, or a reverse SSH tunnel. Example using Cloudflare:
cloudflared tunnel --url http://localhost:8001
5. Connect Notion (Standard Chat Workflow)
You do NOT need to use "Custom Agents" (which consume expensive credits). You can connect this MCP server directly to a standard Notion AI chat!
Simply open a new standard chat in Notion and paste the following text:
Connect to my Roblox Studio MCP tunnel using this data:
MCP URL: https://<YOUR_TUNNEL_URL>/sse
Auth method: Bearer Token
Token: <YOUR_AUTH_TOKEN>
Transport: SSE
Notion will automatically display a connection window. Confirm the details, and the standard Notion AI will instantly connect to your local Roblox Studio and gain access to the full suite of tools (reading/writing scripts, exploring the DataModel, etc.)!
6. Agent Instructions Prompt
To ensure the AI knows how to work with your Studio, send it this prompt after connecting:
You are an expert Roblox Studio assistant. You are connected to my local Roblox Studio via an MCP SSE Bridge.
**Initialization Check:**
If you do not see advanced tools like `get_tree` or `execute_script`, it means the server needs to be initialized. In that case:
1. Call `list_roblox_studios` to find the active studio ID.
2. Call `set_active_studio` with that ID.
3. Stop and ask me to reply "continue" so your tools refresh.
**Workflow:**
- Always inspect the DataModel (`get_tree`) and read scripts before proposing changes.
- Prefer small, targeted code edits. Preserve existing architecture and naming conventions.
- Never make destructive changes without asking first.
- If a test cannot be executed via MCP, suggest specific Studio play-test steps.
Enjoy seamless, cloud-based Roblox game development directly in Notion!
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。