My Plex library has deep cuts. My listening habits don't.

I wanted help choosing what to play from my Plex collection. The obvious homelab response was another app to maintain. Then, I connected my agent to it, because apparently choosing an album needed tooling. That's MusicSeed. My favorite use is asking the agent to find tracks that fit a playlist, reviewing its suggestions and approving the additions. It saves them to Plex; I listen in Plexamp or Cliamp. It exposes the playlist tools through MCP for your agent/harness. There's a web UI and CLI too. Recommendations use your own library's metadata, Plex sonic analysis and listening history. Deployment: Python + SQLite on macOS/Linux. Python 3.12+ required; Node/npm builds the web UI. Import your Plex library first, then configure the MCP connection. The guide covers local Plex databases, SSH snapshots and copied backups. Agent/model hosting depends on your chosen agent. AI involvement: Codex, pi and OpenCode with OpenAI and DeepSeek models assisted throughout implementation, debugging, tests and documentation. Codex also helped draft this post. If you run Plex on a NAS or another machine, I'd especially like to hear where the setup guide gets confusing source: github.com/dafevara/musicseed