In August 2025, I began an experimental project integrating Claude (Anthropic's LLM) with Discord through direct API calls and custom MCP (Model Context Protocol) servers. The system uses Discord as the primary interface and includes proactive messaging capabilities, enabling unprompted reminders and notifications.
This was the first project in which I leaned heavily into the use of LLMs as coding agents. My main objective was both to see what all the hype was about and also to familiarize myself with the process. After working through unnecessary features and crusty code, I feel that I have reached a point in which I've found a good rhythm in "driving" LLMs for both design and coding. My main takeaway, and hopefully the takeaway of other developers, is that your ability to clearly communicate is just as important with LLMs as it is with other people. You can't expect to just give it some vague requirements, then kick it off and go grab a sandwich.
Over the course of development, I created several custom MCP servers to extend Claude's capabilities:
- Atlassian MCP (Jira and Confluence): Used for drafting project tickets and documentation throughout development.
- Google Calendar and Email MCP: Enables automated calendar management and email tasks.
- YouTube MCP: Provides video content discovery and information retrieval.
- Spotify MCP: Allows automated playlist creation and music discovery.
- DateTime MCP: Ensures accurate time and date information for Claude's responses.
- Discord MCP: Provides tools for reading Discord channel history and message data.
- Chromium MCP : Replicates web search functionality for research and information gathering.
- Fitbit MCP: Initially a custom SQL Server health database for calorie tracking, sleep, and mood data which I later migrated to integrate directly with Fitbit's API for automated health data management and photo-based calorie estimation.
- Geolocation MCP: Uses Tasker to send location data from my phone to a webhook, enabling location-aware proactive messaging.
This project has been both enjoyable and educational, offering hands-on experience with cutting-edge AI integration. I plan to continue exploring LLM capabilities and building additional tooling.
General Skills and Experience
- Directing LLMs as coding and design agents while maintaining code quality
- Utilizing Atlassian (Jira, Confluence) MCP Server instances
- Working with Discord bots and applications
- Managing Claude Code configuration, including conversation histories, token usage, agent memory, and sub-agents
Technical Skills and Experience
- Working with the Discord API to send messages, embeds, files, and comment replies
- Working with Claude Code, Claude CLI, and Claude API as LLM interfaces
- Creating custom MCP servers to extend LLM functionality
- Using SQL Server Azure for database backing