Every year around April I end up scrolling through twelve months of transactions trying to piece together numbers for my tax return. This year I had Claude connected to matcha money via MCP, so I just… asked it.

I asked for my charitable giving totals and Claude pulled them by searching across transactions and pulling out the organizations I gave to. I asked for rental property expenses for Schedule E and it totaled costs across multiple utility providers and Zelle payments by merchant name, broken out by category. What was surprising was how well it did this even though I'd never categorized any of these transactions beforehand!

The one that surprised me most: I took a property inspection trip to Pittsburgh last year and needed to document the expenses. Claude searched for transactions in that area during that date range and came back with the hotel, meals, parking, and fuel stops. I'd forgotten about some of those. None of them were tagged or categorized as trip-related, it just found them by location.

I also had it search "interest" across all my connected accounts filtered to 2025 — turns out the Chase checking interest was a few cents and not worth thinking about, but the Vio Bank amount was real. Normally I'd be logging into two different bank sites for that. It listed every check written from checking in 2025 and surfaced a recurring payment pattern I hadn't noticed. And when I asked about 529 contributions, it searched all my accounts and confirmed they weren't flowing through anything tracked — which saved me from digging through the wrong statements.

Here's a taste of what Claude came back with:

Question	Answer
Charitable giving	$841 total
Schedule E expenses	$5,903
Pittsburgh trip	$857

All of this happened in conversation. I asked questions, Claude queried matcha money, and I got back answers with actual dollar amounts. No exports, no spreadsheets. The MCP integration that already existed made all of this possible — nothing here required building anything new.

If you haven't connected your AI assistant yet, here's how.

Chris Hsu