weppy-railway
MCP server for Roblox Studio integration, enabling AI agents to interact with Roblox through SSE transport and a local HTTP bridge for plugin communication.
README
WEPPY Roblox MCP — Railway Host
Deploy @weppy/roblox-mcp ke Railway supaya AI agent (Hermes/Claude/Cursor) bisa connect via SSE, dan Studio plugin bisa hit HTTP bridge lewat reverse proxy 1 port.
Arsitektur
┌─────────────────────────────────────────┐
│ Railway ($PORT) │
Internet ─────────►│ reverse proxy (start.mjs) │
│ │
│ /sse /message /healthz ──► supergateway│
│ │ │
│ └─stdio─► @weppy/roblox-mcp │
│ │ │
│ /* (dashboard, │ │
│ /status, plugin) ──┘ HTTP :3002 │
└─────────────────────────────────────────┘
| Endpoint | Fungsi |
|---|---|
GET / atau /railway-health |
Health + daftar endpoint |
GET /sse + POST /message |
MCP transport (AI client) |
GET /status |
Status WEPPY bridge |
GET /dashboard |
WEPPY web dashboard |
/* |
Plugin HTTP API (poll commands, dll) |
Batasan KRITIS (baca dulu)
WEPPY dirancang localhost:
- MCP transport default = stdio (AI app spawn process lokal).
- HTTP bridge default =
127.0.0.1:3002(Studio plugin di mesin yang sama). - Official README: "The server runs on localhost only (127.0.0.1:3002)."
Jadi Railway saja TIDAK cukup biar AI remote + Studio remote jalan 100% out-of-the-box.
Setup yang realistis
| Skenario | Cara |
|---|---|
| A. Recommended | WEPPY jalan di Windows VPS / PC yang ada Roblox Studio. AI (Hermes di Linux) connect lewat tunnel SSE. |
| B. Full Railway | Deploy ini ke Railway. Studio di Windows connect plugin ke URL Railway (custom host / hosts-file hack / reverse tunnel). |
| C. Hybrid | Studio + WEPPY lokal di Windows. Expose cuma MCP SSE ke internet via cloudflared/ngrok. |
Deploy ke Railway
1. Login CLI
railway login
# atau non-interactive:
railway login --browserless
# paste token dari https://railway.app/account/tokens
2. Init + deploy dari folder ini
cd /home/rafacorps/weppy-railway
railway init # create project "weppy-mcp"
railway up # build + deploy
railway domain # generate public URL
3. Env vars (Railway dashboard atau CLI)
railway variables set MCP_AUTH_TOKEN="$(openssl rand -hex 24)"
railway variables set LOG_LEVEL=info
railway variables set DASHBOARD_AUTO_OPEN=false
railway variables set SKIP_PLUGIN_INSTALL=true
railway variables set WEPPY_MCP_DETACHED_LIFECYCLE=true
# setelah domain ada:
railway variables set PUBLIC_BASE_URL="https://YOUR-APP.up.railway.app"
4. Cek health
curl -s https://YOUR-APP.up.railway.app/railway-health | jq
curl -s https://YOUR-APP.up.railway.app/status
curl -s https://YOUR-APP.up.railway.app/healthz
Connect AI client (Hermes / Claude / Cursor)
SSE (supergateway style)
{
"mcpServers": {
"weppy-roblox": {
"url": "https://YOUR-APP.up.railway.app/sse",
"headers": {
"Authorization": "Bearer YOUR_MCP_AUTH_TOKEN"
}
}
}
}
Via lokal stdio wrapper (kalau client cuma support stdio)
npx -y supergateway \
--sse "https://YOUR-APP.up.railway.app/sse" \
--oauth2Bearer "YOUR_MCP_AUTH_TOKEN"
Config MCP:
{
"mcpServers": {
"weppy-roblox": {
"command": "npx",
"args": [
"-y", "supergateway",
"--sse", "https://YOUR-APP.up.railway.app/sse",
"--oauth2Bearer", "YOUR_MCP_AUTH_TOKEN"
]
}
}
}
Connect Roblox Studio plugin
Plugin default hit http://127.0.0.1:3002. Opsi:
Opsi 1 — reverse tunnel di mesin Studio (paling clean)
Di Windows (Studio machine):
# cloudflared tunnel ke Railway bridge path, ATAU
# jalankan WEPPY lokal (recommended):
npx -y @weppy/roblox-mcp@latest
Kalau WEPPY lokal, AI remote connect via:
# di Windows expose port 3002 + stdio via cloudflared
cloudflared tunnel --url http://127.0.0.1:8787
Opsi 2 — plugin ke public Railway URL
Kalau plugin punya field Server Host / Custom URL di settings:
https://YOUR-APP.up.railway.app
Kalau tidak ada custom host (banyak build hardcode localhost):
# Windows hosts + local reverse proxy (Caddy/nginx) ke Railway
# atau SSH reverse tunnel:
ssh -R 3002:YOUR-APP.up.railway.app:443 user@studio-machine
Lebih gampang: jalanin WEPPY di mesin Studio, Railway cuma buat AI SSE gateway yang proxy ke situ (client-mode) — butuh arsitektur 2-node.
Opsi 3 — WEPPY di Windows VPS (paling stabil buat Valgorant)
[Roblox Studio + Plugin] ←localhost:3002→ [WEPPY di Windows VPS]
↑ stdio
[supergateway :8787]
↑
[cloudflared / ngrok]
↑
[Hermes di Linux VPS]
Script Windows (PowerShell):
# install
npm i -g @weppy/roblox-mcp supergateway
# terminal 1 — expose MCP as SSE
npx supergateway `
--stdio "npx -y @weppy/roblox-mcp@latest" `
--port 8787 `
--baseUrl http://127.0.0.1:8787 `
--ssePath /sse --messagePath /message --cors
# terminal 2 — public tunnel
cloudflared tunnel --url http://127.0.0.1:8787
Lalu di Hermes pakai URL cloudflared /sse.
File di repo ini
| File | Fungsi |
|---|---|
start.mjs |
Reverse proxy + spawn supergateway+WEPPY |
Dockerfile |
Image Node 20 |
railway.toml / railway.json |
Config Railway |
package.json |
deps: weppy + supergateway + http-proxy |
.env.example |
Contoh env |
Local test (sebelum deploy)
cd /home/rafacorps/weppy-railway
npm install
PORT=8080 MCP_AUTH_TOKEN=test node start.mjs
# other terminal:
curl -s http://127.0.0.1:8080/railway-health
curl -s http://127.0.0.1:8080/status
curl -s http://127.0.0.1:8080/healthz
Troubleshooting
| Gejala | Fix |
|---|---|
| Container crash loop | Cek logs: railway logs. Pastikan WEPPY_MCP_DETACHED_LIFECYCLE=true |
/sse 502 |
Supergateway belum ready — tunggu 30–60s cold start |
| Plugin "Disconnected" | Plugin masih ke localhost. Pakai opsi Studio di atas |
| Dashboard blank | Buka /dashboard lewat public URL; hard refresh |
| 401 on /sse | Kirim Authorization: Bearer $MCP_AUTH_TOKEN |
| License / Pro tools fail | Normal di Basic; Pro key diaktifkan lewat plugin UI |
Ringkas rekomendasi buat lu (Valgorant + Hermes)
- Roblox Studio → Windows VPS
104.207.93.128(atau PC lu) - WEPPY + plugin → install di mesin yang sama dengan Studio
- Expose MCP SSE → cloudflared/ngrok dari Windows
- Hermes (Linux) → connect ke URL tunnel
/sse - Railway (opsional) → ganti cloudflared biar URL stabil + auth token
Railway bagus buat URL stabil + auth, tapi Studio harus tetap “dekat” dengan HTTP bridge WEPPY.
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。