Caution

This project is currently under development.

PonyCCG

A full-stack collectible card game engine and ecosystem built around the My Little Pony: CCG ruleset.

Overview

PonyCCG is a Discord-native card game platform featuring automated card ingestion, deckbuilding, and planned gameplay systems. It is an attempt at transforming unstructured game logic into a structured ruleset using LLM-assisted automation.

Card data pipeline

One of the core challenges was converting existing card databases into a structured ruleset that can be usable by a game engine. In order to address this, a simple LLM two-step processing pipeline was created to turn metadata scraped from the amazing dataset from Ponyhead.

Using Cheerio, cards were scraped one-by-one and ingressed into a PostgreSQL database as structured rows via Drizzle. [source]

For each card, the extracted human-readable card ability text is then processed through a transformation pipeline using structured outputs with gpt-5.4-nano, transforming it into a validated JSON representation of game mechanics, including triggers, costs, zones, and operations. Sample of the output of this transformation below:

{
    "name": "Twilight Sparkle, Gala Greeter",
    "start_side": {
        "home_limit": 3,
        "keywords": [],
        "abilities": [
            {
                "trigger": {
                    "event": "card_played",
                    "card_type": "event",
                },
                "operations": [
                    {
                        "op": "flip",
                        "target": "self",
                    },
                    {
                        "op": "exhaust",
                        "target": "self",
                    }
                ]
            }
        ]
    },
    "boosted_side": {
        "home_limit": 4,
        "keywords": ["Studious"],
        "abilities": [
            {
                "trigger": {
                    "event": "card_moved",
                    "to_zone": "problem"
                },
                "operations": [
                    {
                        "op": "move",
                        "target": "opponent",
                        "destination_zone": "problem",
                        "filter": {
                            "card_type": "friend",
                            "zone": "problem"
                        }
                    }
                ]
            }
        ]
    }
}

Compared to the original card page

Zod schemas are used to minimize hallucinated mechanics and ensure generated data remains compatible with the game engine. [source]

Deckbuilding

Using the web interface hosted in web/, users can authenticate via Discord OAuth2 and build a MLP:CCG deck.

Running it yourself

Before running any scripts, ensure that you have correctly configured your working directory.

npm install

# ingress destination
export DATABASE_URL=<your-url-here>

# if using LLM transformation
export OPENAI_KEY=<your-key-here>

Ponyhead Scraper

npm run scraper:run

Scrapes all the cards on Ponyhead and inserts them into your database at DATABASE_URL

LLM transformation

npm run llm:infer

Turns the card text from the cards in your database into actual gameplay rules by processing them through gpt-5.4-nano. Rulesets are then stored in the ruleSet JSON column.

Future project goals

PonyCCG is not currently being maintained as of July 2026. In the future, I plan to have the following features completed end-to-end:

• Nuxt frontend completed
• Playable game with the Discord bot
• On-demand SVG renderer for game canvas