Learn
What is MCP? How an AI assistant connects to your money, in plain English
By the My AI Fin App team · Updated October 3, 2026 · 5 min read
MCP is the reason an AI assistant can answer 'will I run short before payday?' with your real numbers. It is a standard way for assistants to use outside tools. You do not need to be technical to understand what it does or to judge whether a connector is safe.
The short version
On its own, an AI assistant knows only what it learned in training and what you type. It has no idea what is in your bank account, your calendar or your inbox.
The Model Context Protocol, MCP for short, is an open standard for giving an assistant tools. A service that wants to be usable from an assistant publishes an MCP server, often called a connector. You add the connector to your assistant, and from then on the assistant can call the tools it offers.
A common comparison is a universal plug. Before a standard existed, every assistant needed its own custom integration for every service. With one, a service builds a connector once and any assistant that speaks MCP can use it.
What a connector offers
A connector is mostly a list of tools, each with a name and a description the assistant can read. A finance connector might offer 'list debts', 'search transactions' and 'estimate payoff date'. Each tool takes a few inputs and returns a small, structured answer.
You never pick a tool yourself. You ask a question in ordinary words, and the assistant reads the descriptions and decides which tools would answer it.
How one question travels
The whole exchange takes a second or two, and many assistants show you which tool was called so you can see where an answer came from.
- You ask: how much did I spend on eating out last month?
- The assistant sees that the finance connector has a spending summary tool and calls it, asking for last month, grouped by category.
- The connector checks that the request carries a valid token for your account, runs the query with your permissions, and returns the totals.
- The assistant reads the result and answers you in a sentence, often with a comparison or a follow-up suggestion.
Who is allowed to connect
A connector to personal data has to know who is asking. There are two common ways. The first is the familiar sign-in flow: the assistant sends you to the service's own website, you sign in there and choose Allow, and the assistant receives a token. Your password is never shown to the assistant.
The second is a personal access token: a long secret you create in the service's settings and paste into the assistant's configuration. It is more manual, and it suits developer tools that do not support the sign-in flow.
Either way, the token can be revoked from the service's settings without changing your password.
What to check before you add any connector
- Who runs it. A connector published by the service itself is a different thing from one written by a stranger that asks for your password.
- What it can change. Reading is low risk. Tools that send, delete or pay deserve a much closer look.
- Whether you can limit it. Look for a read-only option.
- How you turn it off, and whether turning it off takes effect straight away.
- What is logged about your use of it.
The My AI Fin App connector
My AI Fin App publishes its own MCP connector. It works with Claude, ChatGPT, Cursor and other MCP clients, and offers tools for your overview, accounts and debts, transactions, spending, subscriptions, upcoming bills, budgets, goals, net worth history, payoff estimates and the cash-flow projection.
Connections made through the sign-in flow are always read-only. Every request runs as you under the same database rules as the app, we store only a hash of each personal access token and never the token itself, and you can disconnect any assistant under Settings, Assistants.