Fluduro | A ChatGPT App Project

Fluduro is a simple personality-to-flower quiz application. This project was built to explore the OpenAI Apps SDK and the Model Context Protocol (MCP) by creating interactive widgets directly inside ChatGPT.

🛠️ Project Overview

The goal of this project was to learn how to:

1. Build an MCP Server that exposes tools and UI resources.

2. Use the OpenAI Apps SDK to render custom HTML/CSS widgets in ChatGPT.

3. Connect a local server to ChatGPT using ngrok.

4. Use an LLM (Groq) to dynamically generate quiz questions.

Tools & Resources:

• Tools: say_hello, start, submit_answers, show_results, get_quiz_state.

• UI Widgets: Welcome, Quiz, and Results screens.

Related MCP server: Hello Widget Example

📸 Visuals

![Welcome Screen (Light Mode)](./screenshots/welcome-light.png)
<!-- slide -->
![Welcome Screen (Dark Mode)](screenshots/welcome-dark.png)
<!-- slide -->
![Quiz Screen](screenshots/quiz.png)
<!-- slide -->
![Results Screen](screenshots/result.png)

⚡ Local Setup

1. Installation

Clone the repository and install dependencies:

git clone <repository-url>
cd plantora
npm install

2. Configuration

Create a .env file for your API keys:

cp .env.example .env

Add your Groq API Key to the API_KEY field. You can get one for free at console.groq.com.

3. Run and Tunnel

Start the server in HTTP mode and use ngrok to create a public HTTPS tunnel (required by ChatGPT):

# Terminal 1: Start the server
npm run start:http

# Terminal 2: Start ngrok on the same port
ngrok http 3553

Take note of the public URL provided by ngrok (e.g., https://xyz.ngrok-free.dev).

🔗 How to Connect to ChatGPT

To test this app in ChatGPT, follow these steps:

1. Enable Developer Mode in ChatGPT (Settings → Apps & Connectors → Advanced settings).

2. In Settings → Connectors, click Create.

3. Select Streamable HTTP and paste your ngrok URL with the /mcp path: https://your-id.ngrok-free.dev/mcp

4. Name it "Fluduro" and click Create.

5. Add the connector to a new chat and type "Start the quiz."

🤖 Tech Stack

• Backend: Node.js with @modelcontextprotocol/sdk.

• LLM: Groq (Llama 3.3 70B) for generating questions and analyzing traits.

• Frontend: Vanilla HTML/CSS with Tailwind CSS (CDN) and Google Fonts.

• Tunneling: ngrok.

📖 Key Learnings from the SDK

• window.openai Bridge: Learned how to communicate between the widget iframe and the host.

• Theme Sync: Used openai:set_globals to make the UI adapt to ChatGPT's Dark/Light mode.

• Tool-Driven UI: Learned how to trigger UI changes from the model's tool outputs using structuredContent.

Conclusion

This project served as a hands-on introduction to building native-feeling apps for the ChatGPT ecosystem. It focuses on the basics of tool registration, resource handling, and state management within the OpenAI Apps framework.

📸 Visuals - Screenshots For References

WELCOME SCREEN -LIGHT MODE

WELCOME SCREEN -DARK MODE

QUIZ SCREEN

RESULT SCREEN