How Stocklake Started: From a Google Sheet to an MCP Server
I'm Michael, the person building and running Stocklake. This is the first post in what'll be a series about what Stocklake actually is, how it's built, and what I've learned running an AI-driven trading pipeline on my own money. Figured the right place to start is: where did this even come from?
A Google Sheet, ten years ago
Stocklake didn't start as a company idea or a product idea. It started as a Google Sheet.
Ten years ago I just wanted to track my own portfolio — what I owned, what it was worth, how it moved. I built out some formulas to pull in prices, added a few charts, and it worked. Good enough that I kept expanding it: watchlists, more portfolios, more detail. For years, that spreadsheet was the whole system.
The pandemic rebuild
When the pandemic hit, I had a lot more time to think about the setup, and the spreadsheet was starting to show its limits. I wanted real tracking — proper portfolio value and performance history, some basic alerts, dividend and income tracking. That meant it was time to stop bolting things onto a sheet and actually put the data somewhere real: a database.
That rebuild worked well enough that I didn't stop there. Over the following months I kept adding to it — pulling in news, adding new metrics, wiring up more data sources. Every addition made the next one easier, so the "datalake" (I was already calling it that, informally, in my head) kept growing. What started as a portfolio tracker for one person turned into something a lot bigger than I'd originally planned.
The almost-shipped app
At some point along the way, I started prototyping an actual portfolio tracking app on top of all this — something other people could use, not just me. The first prototype looked genuinely promising.
Then I looked at what else was out there. The big names. The apps everyone already has, or can get in five minutes. And I asked myself the obvious question: why ship another portfolio tracker? If your bank isn't enough, there's already a stack of existing solutions — you connect your accounts, you fight with them a bit, you get most of what you want, maybe you pay for it. It's a crowded, solved-enough problem. I didn't see a reason to add one more app to that pile.
So the app sat mostly finished and unpublished. But the datalake underneath it kept getting better.
Then AI happened
2025 changed the calculus. The question stopped being "should I ship a portfolio app" and became: why not turn the datalake itself into something AI can use directly?
That's the idea that became Stocklake: instead of building yet another frontend, offer an MCP server on top of the data I'd already spent ten years accumulating. But an MCP server for just my own portfolios and watchlists is a niche tool for one person. To make it actually useful — to have a real moat against every other data API out there — I had to extend it beyond my own holdings to cover the broader stock universe too.
And the "connect it to AI" part wasn't new to me by then — I'd already been doing it privately for a while. I had my portfolios and watchlists reviewed weekly by AI: get the broad market context, get the portfolio-level context, then loop through every single position and get an honest read on it. Everything got saved back into Stocklake, and over time that built up a real track record of insight — not just "here's your P&L" but "here's what an AI thought about this position three weeks ago, and here's what actually happened."
Naming it
Coming out of the abandoned portfolio-app phase, I had infrastructure and no name for any of it. Stocks. Datalake. Put them together: Stocklake.
That's where the current chapter starts — building an MCP server, meant mainly for Claude, that agents could actually use. And almost immediately I turned it on myself: connected my own tool to my own agents, on Claude web and desktop, and started using it daily for real analysis. Once I was living in it every day and it was actually useful, the obvious next thought was: if I like this, other people building with AI probably would too.
So it started as just the MCP server. How it grew from there — that's the next post.