Video · GitHub ·
A pixel dog that writes songs: lessons from my first MCP server with GitHub Copilot
In short
Yunika Liu built Coco, a pixel-art golden retriever that runs in a desktop window and Pixoo LED display, using GitHub Copilot and an MCP server to synchronize animations, manage state, and compose music. The project shows how AI coding agents can connect to custom hardware while highlighting practical lessons about incremental experimentation, debugging small details, and designing better memory across agents.
Key points
- Coco supports interactions, notes, moods, timers, and music composition.
- The MCP server lets multiple AI coding agents use Coco’s shared tools and state.
- WebSocket synchronization connects Coco’s desktop version with the Pixoo display.
- Coco can trigger music directly or let Copilot decide when to celebrate task progress.
- Conversation history is not automatically synchronized between AI agents.
- The next plans are DJ Coco with lo-fi, dance, and relaxing music, plus a Stack-chan robotic body.
- A small URL slash caused a major-looking connection problem, reinforcing the value of checking simple details first.
Written by an AI model from the video’s captions and description.
It’s 2 a.m. You’re deep in a solo coding session, and your IDE doesn’t notice you haven’t moved in an hour, but the golden retriever on your desk does.
In this story
A tool or model opens in the Heidelberg AI Curriculum: what it does, what it costs, how to start. A repo opens on GitHub.


