Connect Any AI to Your Finances
Yesterday I wrote about whether it is safe to connect your bank to ChatGPT. My honest take was that the connection itself is about as safe as any account link you have already set up, and that the harder question is who ends up holding your financial life. The belief I kept coming back to was this: the AI reading your finances should be swappable, and your data should be portable. I said Model Context Protocol support was on my roadmap, and that I would rather say so plainly than oversell it.
Today it ships. Trupocket now runs a Model Context Protocol server, so you can connect the AI assistant of your choice to your own finances and ask questions in plain language.
What Model Context Protocol actually is
Model Context Protocol, or MCP, is an open standard for connecting AI assistants to outside tools and data. Think of it as a common plug. Instead of every app inventing its own way to talk to every AI, a service exposes an MCP server once, and any assistant that speaks MCP can use it. Claude, ChatGPT, and Gemini all speak it, and a growing list of other clients do too.
For you, that means the assistant reading your money is one you pick. If you start with Claude and later want to switch, your finances do not move and your connection does not get rebuilt from scratch. You point a different MCP client at the same server and carry on.
What shipped today
The server went live as part of the Trupocket API. Connecting takes one click. You authorize your assistant through a standard "Connect" flow, the same kind of button you have used to link apps before, and you stay in control of that access.
Once connected, your assistant gets a read-only tool suite over your financial data. It can pull your transactions, accounts, categories, payees, hashtags, and scheduled transactions. It can also run the same reports the web app does, such as your financial snapshot, spending and top spending, net worth, monthly cashflow, the forecast, budget spending, and loan and credit card amortization.
So you can ask things like "how much did I spend on groceries last month," "what is my net worth right now," or "which scheduled transactions hit next week," and get an answer grounded in your real numbers instead of a guess.
Reading first, writing next
This first version reads your data. It does not write. Your assistant can look at everything I listed above, but it cannot add a transaction, change a budget, or move anything around through MCP yet.
That "yet" is the point. Reading is the foundation, and it is the half that makes an assistant genuinely useful on day one: ask a question, get an answer grounded in your real numbers. Writing, where an assistant acts on your finances instead of just describing them, is the harder half, and it is what I am building toward next. The full Trupocket API already reads and writes for the integrations you build yourself, so the platform can do this today. Bringing it to an AI assistant is the part I want to get right rather than rush.
Why this is the point of yesterday's post
The OpenAI launch I wrote about puts your whole financial picture inside one company's assistant, with no simple way to bring a different model or take your history elsewhere. My objection was never that AI and money do not mix. They do. The problem is the lock-in.
MCP is the answer I believe in. Your data lives in Trupocket, reachable through a public API with 60+ endpoints that the web app itself runs on. The AI you point at it is your choice, and swapping it is a one-click change rather than a migration. You are not renting access to your own financial history through whichever assistant happens to store it.
That is the whole idea. Own your data, then bring your own AI to it.
Try it
The MCP server is part of Trupocket Premium, which runs $2.99 a month. If you are already on Premium, it is ready now. Connecting it in your assistant is a one-click step, and from there you can start asking questions. If you are new, you can create an account and start tracking your money today, then upgrade to Premium when you want to point an AI at your data. The API docs are public if you would rather see how the connection works before you commit to anything.