What Is Quietly Getting More Expensive: Subscription Price Increases
Two things went live inside Trupocket in August and I never announced either one. Trupocket now finds your recurring charges on its own, and it flags the ones that have quietly gotten more expensive. That is awkward, because three weeks ago I wrote about AI subscriptions turning into a household utility bill and said that none of that is automatic in the app. "In the app" was doing real work in that sentence, and it still holds. What changed is a layer underneath.
The bank promise from that post is staying exactly where it is. Nothing reaches into your accounts, and that is a design choice I intend to keep.
It reads your own ledger
The analysis runs on transactions you already entered, so it needs no bank connection. It walks back through your history, groups charges by who you paid, looks for a rhythm in the dates and the amounts, and returns what looks recurring along with a confidence score and the exact window it examined.
That window matters more than it sounds like it should. The lookback never reaches further than 24 months on any plan, and it shrinks to a shorter span on plans with less history. Free keeps the most recent 90 days. A charge also has to appear at least three times before it counts as recurring at all, so on Free a monthly subscription clears the bar with nothing to spare. That's why a short list on a short window tells you how far back it could see, and why the dates it analyzed come back with every answer.
It also skips what you've already handled. If you set up a Scheduled Transaction for a charge, the suggestion list leaves that payee alone instead of offering to tell you about a subscription you clearly know about. Price watching does the opposite and keeps watching it, because a charge you scheduled is the one worth catching a rise on.
The price increases almost nobody tracks
When I audited every subscription I pay for in July, the list held no surprises. I could have written most of it from memory. The total was the number that actually landed.
There is a layer under the total, though, and it moves without asking. Say a streaming plan goes from $15.99 to $17.99. A cloud storage tier picks up a dollar. The charge still looks like the charge, it still lands on the same day, and your eye slides right over it because you're checking that it happened rather than checking what it cost. Two dollars is not worth a decision. That is the whole trick.
So it reports every recent rise with the old price, the new price, the increase in dollars and percent, the date it took effect, and what that rise costs over a year at that billing frequency. That last number is the one that can change a decision. Here is a hypothetical month: a streaming plan up $2, a cloud tier up $1, a phone plan up $5, a software seat up $4. Each is easy to wave off on its own. Annualized, the four come to $144 a year, and $144 is worth a decision.
That yearly figure is also how the list sorts itself. A 40% rise on a $2 app is about $10 a year, while a 6% rise on a $180 monthly insurance premium is about $130. Sorting by percentage would put the small one on top every time.
You also get your total monthly recurring spend, and how far that total has drifted over the last 6 and 12 months, or over the shorter spans your plan's history allows. Drift understates on short history, because a charge that hadn't started yet when the comparison window opened falls outside it, which pushes the number down and never up.
What it deliberately does not do
It only reports increases. One charge that lands below the usual price is a credit or a refund or a month you skipped an add-on far more often than it is a genuine price cut, so a single low charge never becomes news on its own. A cut that repeats does show up, in the charge's price history and in the drift number.
A rise also drops off the list once you've lived with it for roughly three billing cycles. An increase from last spring is your normal price now, and a feed that keeps reciting old news becomes a feed you learn to scroll past.
What counts as an increase depends on the charge. A contracted subscription that bills the same amount every month gets a tight tolerance, and a metered bill that swings all year gets a wide one, so a variable utility behaving like a variable utility never trips the alarm.
Where you can actually get at this
There are two routes today, and both need an introduction. The first is the API, a documented set of web addresses any program can call to read the same data the web app shows you.
The second is newer. There is an open standard for connecting AI assistants to outside tools and data. A service exposes its data once, and any assistant that speaks the standard can read it, the way any plug fits any outlet of the right shape. The standard is called Model Context Protocol, MCP for short, and Trupocket has run an MCP server since June. You can point Claude or another assistant at your own finances and ask, in ordinary words, what has gone up lately. The detected charges, the price rises, and the monthly rollup all come back through that one read tool. So the detection is reachable through the API and through an MCP assistant.
MCP read access requires a paid plan, which starts at Premium for $2.99 a month. The recurring endpoint on the API is open on every plan, subject to the Free plan's 20 calls a day and its 90-day window. And there is no web app screen for any of it yet. If you were hoping to open a page and see a list waiting for you, that page doesn't exist today, and I would rather say so here than have you go looking.
Nothing pushes this at you either. You ask, or your assistant asks on your behalf, and the answer comes back. That suits me for a first release. A tool that interrupts you about a $2 change earns an ignore, and I'd rather build the number that holds up when you look it up on purpose.
Why this order
The API and the assistant route came first, and the screen comes last, because the detection is the hard part. Getting the grouping right, deriving a sensible threshold per charge, and refusing to cry wolf on a variable utility took most of August. Drawing a list is a week.
The screen is up next on the roadmap. Until then, the number is reachable, and I'd rather hand you a real one now than a polished page later. If you already track your spending in Trupocket, your history is already the input. The only thing left is to ask for the number.