Work

Agent Fantasy - AI Discord Bot with MCP Integration

Agents
Discord
Python
MCP

A Discord bot powered by OpenAI that demonstrates dynamic tool integration using Model Context Protocol (MCP) for unlimited extensibility.

AI Discord Bot with dynamic tools

Project Background

This project explores how AI agents can control Discord bots through dynamic tool systems and proper prompting. Instead of hard-coding every feature, the bot uses the Model Context Protocol (MCP) to load tools at runtime, making it infinitely extensible.

Key Features

Dynamic Tool System

  • Unlimited extensions via MCP SSE servers
  • Tools automatically converted to OpenAI function schemas
  • Modular architecture for easy management

Built-in Capabilities

  • Web search with deep search for detailed research
  • Document fetching and analysis (PDF, DOCX, HTML, etc.)
  • Image and video generation via Hugging Face
  • User/guild/channel information retrieval
  • RAG-powered content search using embeddings

Discord Integration

  • Rich component rendering (containers, buttons, galleries, sections)
  • Custom markdown-to-component parser
  • Interactive elements with context preservation
  • Conversation history awareness

Technical Architecture

Core Technologies

  • discord.py for Discord API interaction
  • openai for AI model integration (supports Azure OpenAI and standard OpenAI)
  • mcp for Model Context Protocol support
  • huggingface_hub for image/video generation
  • google-custom-search for web searching
  • markitdown for document conversion

Component System The bot uses a custom markdown format that transforms into Discord’s UI components:

  • Containers for organized layouts
  • Action rows with buttons (up to 5 per row)
  • Media galleries for images/videos
  • Select menus for multi-choice interactions
  • Sections with thumbnails or buttons

Tool Decorator Functions are converted to OpenAI tools using a custom decorator that automatically generates JSON schemas from Python type hints:

@tool(
    query="Parameter description"
)
async def search(self, ctx: AIContext, query: str) -> Dict[str, Any]:
    """Tool description for AI"""
    # Implementation

This allows seamless integration of new tools without manual schema definitions and helping speed up development and flexibility.

Usage Example

Setup

# Clone and install dependencies
pip install -r requirements.txt

# Configure environment
cp .env.template .env
# Edit .env with your tokens

# Run the bot
python main.py

Adding Custom MCP Tools

# In .env, add MCP server endpoints
MCP_CUSTOM = "https://your-mcp-server.com/sse"

The bot automatically loads all MCP tools on startup and makes them available to the LLM.

Deep Search Implementation

The bot includes a sophisticated deep search system that:

  1. Executes iterative Google searches
  2. Fetches and processes content using embeddings
  3. Ranks results by similarity to target context
  4. Iterates until quality threshold is met (90% rating, 10+ results)
  5. Generates final summary from relevant context

This combines search, RAG, and LLM summarization for comprehensive research.

Docker Support

# Build image
docker build -t ai-fantasy .

# Run container
docker run -d --name ai-fantasy \
  -e BOT_TOKEN="your_token" \
  -e OPENAI_API_KEY="your_key" \
  ...
  ai-fantasy

Project Impact

This bot demonstrates:

  • Agents can effectively control complex systems with proper tooling
  • MCP enables truly modular Agentic applications
  • Custom component systems enhance user experience beyond plain text
  • RAG improves search quality significantly over basic web scraping

Acknowledgments

Special thanks to Dương Lê Giang for guidance on RAG concepts that formed the foundation of the search implementation.

Repository

Source code available on GitHub: Agent Fantasy

Note: Requires OpenAI API access and Discord bot token. Optional features need Hugging Face token and Google Custom Search credentials.