Novistopheles

Song after song lands perfectly. The vibe is locked in. I’m pedaling harder because the music is that good. By the time I get home, I’m thinking: I want this playlist forever. Every morning. This exact mix.

So I open Spotify to save it and… nothing. There’s no option. DJ X plays you incredible, personalized music—and then it vanishes. You can’t save a DJ session as a playlist. It’s like having the best meal of your life and the restaurant won’t give you the recipe.

Okay, I think. AI got me into this. Maybe AI can get me out.

The ChatGPT Letdown

I pull up ChatGPT, figuring there’s probably a Spotify connector. Sure enough, there is. I link my account, feeling clever, and ask it to grab my recent listening history.

It returns five songs.

Five. Out of a two-hour session.

What the heck? Why does every ChatGPT connector feel like it’s been deliberately hobbled? It’s like being handed a sports car with a governor that caps you at 25 mph.

Hello Claude, My Old Friend

This is where MCPs come in.

MCP stands for Model Context Protocol. Think of it like a USB cable that lets an AI agent plug directly into another service—not through some limited app integration, but through the full API. The real thing.

There are a few Spotify MCP servers floating around. The good ones can pull 50 songs from your history. That’s almost two hours of music. That’s a real listening session.

I run Claude on my Mac—both the desktop app and Claude Code in the terminal. Both can use MCP servers, which means I can start a conversation on my phone and finish it on my desktop with full access to my connected tools. It’s seamless in a way the other AI tools just aren’t.

So I installed a Spotify MCP server, set up a simple skill, and now I have a little workflow that does exactly what I wanted.

How It Works

When I ask Claude to make a playlist from my listening history, here’s what happens:

1. It pulls my recent tracks from Spotify
2. It looks at the timestamps and figures out which songs were part of the same session (gaps between tracks tell the story)
3. It knows the genre, tempo, and vibe of each song from Spotify’s metadata
4. It groups them intelligently and asks: “I see a couple of sessions here—which one do you want to save?”

From there, I can have a conversation. Sort by tempo. Drop that one track that snuck in from my kid’s account. Rename it something memorable. And then: make that playlist for me.

It does. It appears in my Spotify library. Done.

The Setup (It’s Easier Than You Think)

If you’ve got Claude Code running with file system access, you can literally say:

“Research the best Spotify MCP server, recommend one, install it, and update my Claude desktop config so I can use it there too.”

Claude will do the research, make a recommendation, handle the installation, and update your JSON config file. You’ll need to restart Claude (both desktop and Code), but after that—you’re connected.

Here’s a starter prompt once you’re set up:

“Pull my Spotify listening history from today. Identify any distinct playlist sessions based on timing gaps. Show me what you found and let me pick which one to turn into a saved playlist.”

Why This Matters (Sort Of)

Look, this isn’t going to transform your business. It’s not a productivity hack that’ll 10x your output.

But that’s kind of the point.

Not everything about AI agents has to be marketing funnels and process automation and serious work stuff. Sometimes an AI can just… do something useful for you. Something fun. Something that makes your morning bike ride a little better because now you can keep that perfect playlist.

That’s worth something too.

Scott Novis is the founder of GameTruck and spends way too much time making AI do things it probably wasn’t designed to do. He writes about business, technology, and occasionally, bike rides.