What I Learned Building a GA4 MCP Server as a Non-Developer
Spoiler alert: It’s not as magical as everyone makes it sound
Let’s Be Real About My Coding Skills
Okay, let me start with some brutal honesty here. I am NOT a developer. Like, not even close. I can fumble my way through some Python code and understand the basic syntax, but that’s literally it. I’m basically that person who Googles “how to center a div” every single time (don’t judge me).
But here’s the thing—I keep hearing all these wild claims about AI coding tools. “Anyone can build apps now!” “You don’t need to code anymore!” Sound familiar?
Yeah, I was skeptical too. So this weekend, I thought, “You know what? Let me actually test this myself.” I decided to build something real: a Google Analytics 4 MCP server that could hook into Claude Desktop, log into my GA4 account, grab all my properties, and pull some reports.
I went on a little tool tour—tried Lovable.dev, Bolt, and Cursor before. They all had their moments, but honestly? Most of them left me wanting more. Until I got to Claude Code. That’s where things got… interesting.
Holy Moly, This Thing Actually Plans!
Here’s what blew my mind right off the bat: Claude Code didn’t just dive headfirst into writing code like some caffeinated intern. Nope. It actually made a plan.
When I threw my GA4 MCP server idea at it, it came back with this beautifully laid out roadmap that was ambitious but totally doable. Even my non-developer brain was like, “Oh snap, this actually makes sense!”
So we started with the authentication stuff—you know, getting cozy with Google’s APIs and making sure we had the right permissions to peek at my GA4 data. And honestly? It went pretty smoothly! Sure, there were a couple of “uh oh” moments with some error messages, but the second I copy-pasted those errors back to Claude Code, it was like, “Got it, fixed!”
I was genuinely impressed. Maybe this AI coding thing really was the real deal…
And Then… Things went sideways
Now that we were all authenticated and buddy-buddy with Google, it was time to actually pull some data. Claude Code started working on listing all my GA4 properties and accounts. Should be straightforward, right?
Wrong.
Even though the authentication was working perfectly, I kept getting this super annoying “no properties found” error. And here’s where I discovered the “AI confidence syndrome.”
Claude Code’s response every single time? “Oh! I see the issue. I’ve fixed it now!”
Runs the code again.
Same error.
“My bad! Found the real problem this time. It’s definitely fixed now!”
Runs it again.
Same. Freaking. Error.
This happened five or six times. I’m sitting there watching this AI tool confidently tell me it’s solved the problem while making the exact same mistake over and over again. It was like watching someone repeatedly walk into a glass door but insisting they’ve figured out how to open it.
At this point, I was getting pretty frustrated. Welcome to what I like to call “AI loop hell”—where the AI gets stuck making the same logical mistakes while acting like everything’s totally fine.
Plot Twist: I Had to Become Sherlock Holmes
Here’s where my weekend took a wild turn. Remember how I said I barely know how to code? Well, turns out I had to channel my inner senior developer real quick.
I rolled up my sleeves and did some detective work:
- First, I stalked the code Claude Code had written. I mean really looked at it, trying to figure out how it was attempting to talk to the GA4 API
- Then I went down a Google rabbit hole diving into the GA4 API documentation (fun Saturday night, am I right?)
- Eureka moment! I finally spotted the issue: Claude Code was trying to list accounts and properties separately when they actually needed to be handled together.
Once I figured out what the functions were actually supposed to do, I went back to Claude Code and was like, “Hey buddy, here’s what you’re doing wrong, and here’s how to fix it.”
And wouldn’t you know it? Claude Code immediately rewrote the whole approach, and from that moment on, everything worked like a charm. It was honestly pretty satisfying!
The Big Reality Check Nobody Talks About
So here’s the tea: AI coding tools are pretty cool, but they’re not the magical “anyone can build anything” solution that all the hype makes them out to be.
Here’s what I learned the hard way—to actually succeed with AI coding tools, you basically need to wear one of two hats:
- The Product Manager Hat: You gotta be really good at breaking down complex stuff into bite-sized, super specific tasks that the AI won’t get confused by
- The Senior Developer Hat: You need to understand code well enough to spot when the AI is going in circles and be like, “Nah fam, that’s not it. Try this instead.”
Without one of these superpowers, you’re basically going to end up stuck in that AI loop hell I mentioned earlier. You’ll be sitting there watching the AI confidently mess up the same thing over and over while you have zero clue how to help it get unstuck.
And trust me, that’s not a fun place to be on a Saturday afternoon.
What This Actually Means If You’re Like Me
Look, AI coding tools are legit impressive, but they’re definitely not the “coding is now accessible to everyone” miracle that people love to hype up. They’re more like really smart assistants who still need a boss who knows what they’re doing.
If you’re a fellow “vibe coder”, these tools can totally work for:
- Simple apps that aren’t trying to cure cancer
- Internal tools that only you and your team will see (so nobody judges the janky parts)
- Learning projects where you’re just trying to figure stuff out
But here’s the kicker—the more complex your project gets, the more you actually need to understand what’s happening under the hood. Sorry, but that’s just the reality.
So, What’s the Bottom Line?
Here’s the deal: I actually did build my GA4 MCP server over the weekend, and it works! That literally would have been impossible for me without AI help. But let me be crystal clear—it wasn’t the smooth, “just describe what you want and magic happens” experience that you see in all those flashy demos.
The real superpower of AI coding tools isn’t that they replace developers (spoiler alert: they don’t). It’s that they can help bridge those knowledge gaps and speed up your learning—but only if you’re willing to get your hands dirty, dive into documentation when things break, and put on that senior developer thinking cap when the AI gets stuck.
If you’re thinking about jumping into AI-assisted development, here’s my honest advice: go for it, but embrace the fact that you’re still going to need to learn stuff. The AI will write the code, sure, but you’re still going to need to understand what the heck it’s actually doing.
And honestly? That’s probably a good thing. You don’t want to be completely clueless about the code that’s running your projects anyway, right?
Curious to test out the GA4 MCP server I built? Link to Github – And hey, drop a message me on LinkedIn if you’ve had your own adventures (or misadventures) with AI coding tools. I’d love to hear your war stories!