The Kid in the Candy Store Problem
It was a notification.
Just one. Someone, somewhere, used Stacky.
Not me testing it. Not my co-founder poking at the endpoints. A real person, doing a real thing, with something I built.
I stared at the screen longer than I should have.
For context
Stacky is our MCP server for Stacklist.
MCP stands for Model Context Protocol. Anthropic's standard for letting AI assistants talk to external tools. If you've ever connected Claude to Gmail or Google Drive, you've touched MCP.
Stacklist is where businesses and people build hubs of organized, browsable content from across the internet. Cards live inside stacks. Stacks live inside a hub. The hub is the thing someone visits, browses, shares.
Stacky is what happens when you give an AI assistant hands inside that system.
Add cards from a URL. Create a new stack. Search your library. Auto-tag content. Generate a summary across a whole collection. Save markdown as a knowledge capsule. All from a conversation.
What it actually took
Writing the tools was the fast part.
The slow part was everything tools don't show you:
- Auth flows that don't break when the token rotates
- Error messages that actually tell the model what went wrong
- Tool descriptions written for a reader who is not a human
- Pagination that respects the model's context budget
- Input schemas that survive being called wrong
- Rate limiting that is kind to the user but not lazy
I spent more time on tool descriptions than on tool logic. That surprised me. MCP isn't really an API. It's closer to writing a very small onboarding document for every single function, over and over.
If you want to try Stacky with Claude, there's a setup guide and it's free.
The candy store
Here's the part I want to be honest about.
Stacklist has 14,000 people using it. That's not the thing that made me stare at my screen. I've shipped software my whole career. I know what launching feels like.
This was different.
For two years, we've been talking about this. What if you could just talk to your hub? What if the AI could add cards, build stacks, search across everything you've curated? What if the thing you made for humans to browse could also be the thing an AI helps you grow?
Two years of whiteboards. Two years of "one day." Two years of the idea being a shape inside our heads that we kept pointing at with our hands.
And now it's just... there.
Not a slide. Not a diagram. Not a Notion page. A thing. You can open a conversation and talk to it. You can ask it to find something and it finds it. You can ask it to create and it creates. You can ask it to improve a stack and it does.
The sentence in my head is just: wow.
I'm a kid in a candy store.
Not because I've never shipped before. Because this specific thing, the one we kept describing to each other for two years, is alive. And it's nice. And it works. And someone is using it right now.
That loop closing from "what if" to "oh there it is, that's what it feels like". I didn't know it would hit me this hard.
A small honest thing
The strange part of building is how rarely you feel the finish line.
Features ship and disappear into a backlog. Migrations complete and nobody celebrates because the system just works now. Wins go quiet fast.
But some things aren't like that. Some things are two years of talking, and then one day the thing is real, and it's used, and you get to watch it be used.
That's a different kind of quiet.
The good kind.
So that's where I am this week. Watching logs. Fixing what's rough. Grinning at my laptop.
Kid in a candy store.
I'll take it.