Cloudinary MCP App → ChatGPT Workshop

A Streamable HTTP MCP server that uploads images/videos to Cloudinary and exposes a rich interactive UI that works in:

✅ ChatGPT Apps
✅ Goose
✅ Any MCP host

This repo is used in the workshop:

“Plug Your MCP App into ChatGPT”

It demonstrates how to:

• Convert an MCP App into a ChatGPT-compatible app
• Serve MCP over Streamable HTTP
• Host it on Railway
• Use one server for Goose + ChatGPT

---

Architecture Overview

ChatGPT
   ↓
Streamable HTTP (Railway)
   ↓
Cloudinary MCP Server
   ↓
Cloudinary API

The same server supports:

• Goose MCP Apps (ui:// resources)
• ChatGPT Apps SDK (skybridge template + structured output)

---

Features

• Upload images/videos to Cloudinary
• Interactive UI preview
• Works in ChatGPT + Goose
• Structured tool output for ChatGPT
• MCP App UI for Goose
• Remote Streamable HTTP deployment
• Stateless server (safe for hosting)

---

Requirements

• Node.js 18+
• Cloudinary account

Get credentials:

👉 https://console.cloudinary.com/settings/api-keys

You’ll need:

CLOUDINARY_CLOUD_NAME
CLOUDINARY_API_KEY
CLOUDINARY_API_SECRET

---

Local Development

Install

npm install

Run dev server

npm run dev

Server runs at:

http://localhost:3000

Health check:

GET /

MCP endpoint:

POST /mcp

---

Environment Variables

Create a .env file in the project root:

CLOUDINARY_CLOUD_NAME=...
CLOUDINARY_API_KEY=...
CLOUDINARY_API_SECRET=...

Do NOT commit this file.

---

Deploy to Railway

1. Push repo to GitHub
2. Go to https://railway.app
3. New Project → Deploy from GitHub
4. Select this repo
5. Add environment variables in Railway dashboard
6. Deploy

Railway will give you a URL:

https://your-app.up.railway.app

Your MCP endpoint becomes:

https://your-app.up.railway.app/mcp

---

Connect to Goose

Add a remote extension:

goose configure

Choose:

Remote Extension (Streamable HTTP)

Endpoint:

https://your-app.up.railway.app/mcp

Now Goose can call the Cloudinary upload tool.

---

Connect to ChatGPT

In ChatGPT:

Add a custom MCP app using the same endpoint:

https://your-app.up.railway.app/mcp

ChatGPT will:

• load the template UI
• inject structured tool output
• render the upload preview

---

Available Tool

upload

Uploads media to Cloudinary.

Parameters:

• file_path — local path (Goose)
• file — URL or data URI (ChatGPT)
• resource_type — image/video/raw
• public_id
• overwrite
• tags

Returns:

• JSON metadata
• interactive UI
• ChatGPT structured output

---

Workshop Goals

This project teaches:

• MCP server structure
• Streamable HTTP hosting
• ChatGPT adapter pattern
• Template + structuredContent bridge
• Cross-host UI compatibility