Hi Reader -

I was not a programmer. My entire technical background was: ‘hire a developer and hope for the best’.

As of a year and a half ago, my big AI move was exporting a CSV from my MMP, uploading it to ChatGPT, and asking it questions about my performance data.

I thought I was ahead of the curve. I was very proud of myself.

Turns out I was doing pivot tables with a chat interface on top.

Something changed though.

I started writing code with AI. I now spend 35 to 40 hours a week on my Terminal(I know that because I wrote code to track and give me insights on where my screen time goes).

And something interesting has happened over the last few months. Our internal workflows ‘talk’ to each other.

Customer insights, performance metrics, creative briefs: all connected in a single system. And that’s when AI went from a novelty to something actually useful.

Because here’s the thing nobody tells you about AI. Chatting with something that has zero context about your business is like asking a rando on the street for life-advice.

Smart-sounding answers. Completely disconnected from reality. The AI isn’t stupid. It just doesn’t know anything about you.

Connected context changes everything.

Which is what made the AppsFlyer MCP connection so interesting.

​

…and which is why I was genuinely excited when our friends from Appsflyer showed me their MCP earlier this month.

Instead of exporting data and uploading it somewhere, the AI is now talking directly to live performance data. Real numbers. Real channels. Real spend distribution. In real time.

So instead of the morning ritual: open MMP, check each channel, copy numbers into spreadsheet, stare at spreadsheet, wonder what changed, decide it probably matters, do nothing about it: I just ask:

“What’s my ROAS by channel right now?”

“Where should I be moving spend?”

“What changed this week and does it actually matter?”

And I get answers grounded in what’s actually happening. Not because the AI got smarter. Because it finally has something real to look at.

The layer on top that makes it actually work is a file called CLAUDE.md — a simple set of instructions that tells the AI how to think about the data specifically. How to calculate ROAS our way. How to rank channels. When to flag something worth caring about. Without it the connection is just a fancy dashboard. With it, it’s an analyst that knows how your business works.

I built the whole thing in about ten minutes. Without a computer science degree. Without a technical co-founder. Without even particularly knowing what I was doing at the start.

If I can do it, you have absolutely no excuse.

I’ve documented the whole setup: the MCP connection, the CLAUDE.md template, the exact queries, all of it: in a full walkthrough with video. If you want to see exactly how this works and build it yourself, it’s all right here.

​https://www.rocketshiphq.com/appsflyer-mcp-performance-agent-claude-code-2/​

Or don’t. I’m sure uploading last week’s CSV export will get you there eventually.

Later -

PS: Quick heads up: this is a partner collab with AppsFlyer. I’ve worked with them for over a decade, which in marketing years is basically a geological era. I’m genuinely impressed by what they’re building and I wouldn’t be writing this if I wasn’t.