The conversation started at work.
I mentioned offhand that I wished I could review my finances with Claude — not a spreadsheet, not a budgeting app, but an actual conversation about five years of spending with something that understood the context behind the numbers. My manager had the same thought. He hadn't done any homework on it. I had.
The Plaid API wants merchant account developers. I don't qualify. That wall closed the obvious path, so I found a different one: download five years of monthly statements manually from Chime and CashApp, clean them, and build something Claude could actually read and query.
The hardest part was the down. Not the cleaning, not the data modeling, not the MCP server. Just sitting there down file after file, month after month, five years back. That's the work that doesn't look like work from the outside.
PlaidMCP came out of that process — a local service that holds structured transaction history, categorized spending, and trend analysis across the full five-year window. Phase by phase, the same way everything else in the stack gets built. Collect the data. Expose it. Query it.
I went into the analysis looking for places to cut. Frivolous spending, subscriptions I'd forgotten, purchase patterns that didn't match my actual priorities. I expected Claude to find the holes and help me fill them.
It looked at five years of data, looked at what it knew about my life and needs, and told me I was doing rather well given my circumstances. Making it week to week. Earning enough monthly. No excessive lifestyle. No obvious waste.
The problem was the bills schedule. Enough payment clustering in certain weeks to create a recurring deficit of around five hundred dollars. Not a spending problem. A timing and volume problem.
The conclusion it reached: I didn't need to spend less. I needed to make more.
That was the thing I already felt but feared. I'd hoped the data would find a different answer — a set of cuts that would make the math work without requiring anything harder. The data didn't find that answer because it wasn't there. The spending wasn't the problem. The income ceiling was.
I came to the same conclusion I'd been avoiding. Saving more isn't really feasible. The life I've built isn't excessive — it's just priced slightly above what my current income handles comfortably. The path forward is career change, secondary income, or both. Not a spending audit.
PlaidMCP didn't tell me anything I didn't already know. It confirmed the thing I already felt with enough specificity that I couldn't argue with it anymore.
The tool isn't finished. Phase 6 is pending — transfer exclusion, better category granularity. What I actually want is annotation: purchase by purchase context captured close to the time of the purchase. The item, the reason, the hidden costs, the justification that made sense in the moment. A financial journal that happens to be queryable. The current version shows me what I spent. The next version will show me why, and whether it held up.
That version is worth building because the question it answers isn't "where is the waste." It's "what do I actually value when I'm honest about it in real time, and what does that cost me."
Those are different questions. The second one is harder to avoid.
The intake page is there if you're building toward something similar.