01 Founder

Daksh Khanna Founder of CashBFF, the personal finance MCP.

02 Essay

I'm building the personal finance MCP

I've always struggled with my money, whether I had some or not.

When I was broke, it was the obvious kind of stress: am I going to make rent, did that charge clear, can I really afford this coffee. When I had a little more, the stress just changed shape. Now it was: am I saving enough, was that a stupid purchase, why does someone earning what I earn feel this anxious about a $40 dinner.

The apps were supposed to fix this. I tried most of them. Mint, then Monarch, then a spreadsheet, then back to an app, then back to the spreadsheet. They all had the same problem, which is that they were a place I had to go. A tab to open, a dashboard to read, a number to interpret. The whole point of a finance app is to make money less stressful, but every one of them was an extra room in my life I had to walk into. Most days I just didn't.

What I noticed, eventually, is that the only place I actually went every day was a chat box. iMessage. Claude. ChatGPT. My whole life had quietly moved into conversation. When a money question came up I'd ask the chat box first, get something generic back, and then not bother opening the real app to find the real answer.

So I flipped the question. Instead of "how do I build a better finance app," I asked "what if the money coach just lived inside the chat box I'm already in?" That's CashBFF. Not an app you open. A capability you talk to, wherever you already are.

I'm calling it the personal finance MCP. I launched it as one in May 2026.

Why "MCP" is the right word for this

MCP stands for Model Context Protocol. It's an open standard Anthropic introduced in November 2024, now governed by the Linux Foundation's Agentic AI Foundation, and the simplest way to describe it is the one Anthropic uses themselves: a USB-C port for AI. Before USB-C, every device had its own weird cable. MCP does the same thing for the connection between an AI assistant and the real world: your data, your tools, your accounts. Build one MCP server and any AI client that speaks the protocol can plug into it.

A personal finance MCP, then, is that universal port for your money.

Here's what it actually looks like. You're texting with Claude. You ask: "am I going to make rent this month?" Claude doesn't guess. It calls the CashBFF MCP, which looks at your real financial picture (income, spending, what's already committed) and answers you with a number and a reason. You didn't open an app. You didn't pull up a dashboard. You asked your AI a money question and it had the context to actually answer it.

The reason I keep saying "personal finance MCP" instead of "AI money app" is that those two things are different shapes, and the difference matters.

Why an app is the wrong shape

Cleo, Monarch, Copilot Money, Rocket Money. These are good products, genuinely. I'm not here to bury them. But they're all the same shape: an app you install, open, and check.

That shape has three problems.

Every app starts at a loss. Before it can help you, it has to convince you to download it, sign up, link your bank, and come back tomorrow. Most people fall out somewhere in that funnel. A personal finance MCP has no funnel. If you already use Claude, you're one connector away from done.

Apps make you do the work. You open Monarch, you read the dashboard, you draw the conclusion. You're operating the tool. When the AI has the context, the AI does the operating. You just talk.

Apps fight where you actually are. You're not in an app. You're in a thread. Asking "can I afford this?" to the same chat box where you already plan your week, draft your texts, and look stuff up is not a feature, it's the whole game.

Cleo is the closest thing to what I'm building, and it's worth being precise about how we differ. Cleo is a budgeting app that uses an AI-powered chatbot, and it does that well. You DM it like a friend, it roasts you, it connects to your bank through Plaid. But it's still an app. You download Cleo, you live inside Cleo's chat, you trust Cleo's model. CashBFF flips it. CashBFF doesn't bring you to a new chat. It brings your money into the chat you're already in.

That's not a small distinction. That's the entire bet.

How they actually compare

Product Shape Where you live Who does the work
Cleo App with AI chat Inside Cleo Cleo's chatbot
Monarch Money Dashboard A finance "command center" with customizable widgets and net worth tracking You
Copilot Money Dashboard A sleek iOS/Mac app with ML categorization and a chat interface in development Mostly you
Rocket Money Subscription manager + dashboard A beginner-friendly tool that finds and cancels recurring charges App + you
CashBFF MCP The AI you already use The AI does it

Everyone in the top four is competing for the same prize, which is your home-screen attention. I'm not trying to win that fight. I'm trying to skip it.

What CashBFF actually does

If you're technical, here's what's under the hood. An MCP server can expose three kinds of things to an AI client: resources (data it can read), tools (actions it can take), and prompts (templates). CashBFF only uses tools. There are 13 of them today.

Six read tools that let the AI see your money:

Tool What it does
get_my_money_snapshotrunning balance, the next 30 days of recurring stuff, the next 90 days of scheduled stuff, plus recent transactions, all in one blob
get_my_balancescash on hand minus credit-card debt, per account
get_my_upcomingwhat's coming up in the next 60 days
search_my_transactionskeyword and merchant search of your bank transactions
get_my_runwaywhen do I run out
list_recurring_suggestionsrecurring streams the model has detected but you haven't confirmed yet

Seven write tools, gated behind a flag, that let the AI act on your money:

Tool What it does
add_goalshort text like "pay off card by july"
add_transaction_noteoff-bank entry, cash or IOU or side income
add_scheduled_transactionforecasts like "expecting $1,200 rent on the 1st"
edit_scheduled_transaction and delete_scheduled_transactionbut only on rows the user created
mark_transaction_reimbursableflag a charge for later reimbursement
confirm_recurring_suggestionpromote a model-detected recurring stream to confirmed

The one I keep thinking about is get_my_runway. Most finance apps show you what you spent. A pie chart of last month, and you draw your own conclusions. get_my_runway does something different. It looks at your current spend rate, your scheduled income, your upcoming bills, and answers the only question that actually changes behavior, which is "how many days do I have." Not where your money went. Not whether you stayed under budget. Just: at this pace, when do you run out.

It's a short, almost rude tool. It's also the one that makes people pause before the next purchase, and I think that pause is the only thing a finance product can do that matters.

Two more I want to call out because they only really work in a chat: confirm_recurring_suggestion and add_transaction_note. The first is the AI saying "hey, looks like something called RKT-TKNS has been charging you $19 every two weeks, want me to track this as recurring?" That's an awkward tap-driven UI. It's a natural sentence. The second is "I lent Sam $40 in cash on Tuesday, can you note that," which is just a thing you'd say to a human and which no app makes easy.

How you actually connect it

CashBFF is a remote HTTP server. There's no npm package and no token to paste. You sign up at cashbff.com, subscribe through Stripe, link your bank through Plaid, then add CashBFF as a Custom Connector in Claude. You approve a consent screen, the OAuth handshake mints a session JWT, and from that point on Claude can call CashBFF's tools whenever you ask it a money question. Whole thing takes about a minute. You never see a token.

On the data side: Plaid is the only source, access tokens are encrypted at rest with AES-256-GCM, and the server reads far more than it can write. It can see balances, transactions, recurring expenses, scheduled forecasts, and the last four digits of your account masks. It can't move money. It can't see full account or routing numbers. It can only edit things you explicitly created. The default is read.

For what it's worth, I built the first version of CashBFF as an SMS bot during my Prometheus X Venture Fellowship (which came with $5,000 in non-dilutive funding) and Claremont's student accelerator. The MCP is the same idea, ported to a surface that already has the AI built in.

What I'm learning from early users

I want to write this honestly, because the temptation in a founder essay is to fill it with stuff that sounds good. So here's what's actually true.

The first version of CashBFF was an SMS bot with the same kinds of tools the MCP has now. The number one question I saw, by a wide margin, was some version of "can I afford to go get drinks." Almost always sent between 2 and 3 AM. That tells you something about what people actually want from a money product. It's not "build me a budget." It's "help me make this decision in the next forty seconds."

The thing that surprised me most was that a meaningful number of early users mentioned, unprompted, that they'd never been able to stick with a finance app before. Different reasons, same outcome. They downloaded Mint or Rocket Money, used it for a few days, and quietly stopped. The pattern with CashBFF was different. People used it once, came back the next day, and turned into daily users without me asking them to. The wedge of "ask one question" became a habit faster than I expected, and I think it's because asking is lower-friction than logging in.

One user told me that the part she used most was forecasting. She'd never planned future expenses before because no app made it feel natural. With CashBFF she was doing it in a sentence ("expecting a $400 cell bill next Tuesday, plus my Verizon refund the week after") and getting a real answer back about what that meant for the rest of the month. That one stuck with me, because she was discovering she wanted forecasting without anyone teaching her what forecasting was.

Casey Ariel Dike, writing about CashBFF on Interledger.org, put it this way:

A Gen Z founder building CashBFF, an AI-powered text-messaging platform that turns financial coaching into friendly, human conversations. It's simple. It's intimate. And it speaks the language of his generation.

I think "intimate" is the right word. It's also the word that explains why a chat box is the right surface for this and a dashboard isn't. People will tell a chat thing things they would not type into a form. That changes what a finance product can know about you, and therefore what it can do for you.

What's next

CashBFF today is the first version of a bigger idea, which is that your money should be something you can just talk to, through whatever AI you already trust, on whatever surface you already use.

The thing I'm actually trying to build is the fairest-priced version of a tool that helps people who struggle to see and manage their money live their lives more fully. Not a dashboard. Not a coach you have to remember to open. Something that's already there, in the thread, when you have the question. The personal finance MCP is the wedge. More surfaces, deeper context, and the ability to not just answer questions but take action on your behalf, all of that is what's next.

If you're building at the intersection of AI and money, or you just want to argue with me about any of the above, I'd love to hear from you.

And if you'd rather feel this than read about it, try CashBFF.

— Daksh

03 Recognition

Press & affiliations

  • 01

    Rise

    1 of 100 selected from 80,000+ applicants, 2023. Pictured in Teen Vogue.

  • 02

    Prometheus X

    Venture Fellow, Class of 2025 · $5,000 non-dilutive funding

    Daksh is a systems builder and mathematically trained thinker creating emotionally intelligent technologies that restore agency and confidence — especially around money.

    Eddie Mandhry
  • 03

    Interledger Foundation

    Founder profile by Casey Ariel Dike

  • 04

    NCDC (JAG)

    Judge, 2 student competitions · CashBFF expo host

  • 05

    AFROTECH

    Attendee