Overview

pokebuilder is an AI-powered application designed to assist users in building effective Pokemon Trading Card Game (TCG) decks. It leverages advanced AI models and a vector database to analyze regulations, card data, and user queries to generate optimized decks.

The application consists of the following components:

• Crawling Module: Fetches and processes data from online sources, such as regulations and card details.
• Vector Database: Stores embeddings of regulations and card data for efficient similarity searches.
• MCP Server: Provides an API interface for generating decks based on user queries.
• RAG (Retrieval-Augmented Generation): Combines retrieved context with user input to generate meaningful prompts for deck building.

Features

• Crawls Pokemon TCG regulations and card data from specified URLs.
• Stores processed data in a Milvus vector database for fast retrieval.
• Uses AI embeddings to match user queries with relevant data.
• Generates optimized Pokemon TCG decks based on user input and contextual data.

Prerequisites

Before running the application, ensure you have the following installed:

• Python 3.10+
• A virtual environment tool (e.g., venv or virtualenv)
• Milvus vector database
• Required Python dependencies (listed in requirements.txt)
• Local Ollama installation
• Google Gemini API key

Installation

1. Clone the repository:

       git clone https://github.com/your-repo/pokebuilder.git
       cd pokebuilder

2. Create and activate a virtual environment:

       python3 -m venv .venv
       source .venv/bin/activate

3. Install dependencies

       pip install -r requirements.txt

Usage

1. Crawling data Run the crawling script to fetch and process regulations and card data:

       python3 src/crawl.py

   This script will:

   • Crawl URLs listed in crawl/urls/regulations/urls.txt and crawl/urls/tactical_decks/urls.txt.
   • Process the content and store embeddings in the Milvus vector database.

2. Starting the MCP server Start the MCP server to handle user queries and generate decks:

       python3 src/mcp_server/py

   The server will run on the host and port specified in the .env file (default: http://0.0.0.0:8051).

3. Querying the server For testing purpose run the MCP inspector using following command:

       mcp dev src/mcp_server.py

   Within the inspector, send a query like:

   Build a deck with PikachuEX and CharizardEX as the main cards with 2 energy type cards. The server will return an optimized deck based on the provided query and contextual data.