What is an MCP server?
You have reached the advanced module. This lesson explains the technology that makes “connect my AI to my portfolio” possible, in plain language. No coding required.
The problem it solves
Section titled “The problem it solves”An AI assistant on its own is a closed box. It has language ability and general knowledge, and no way to reach anything else. It cannot read your calendar, look at your files, or see your portfolio.
Historically, every combination of assistant and service needed its own custom integration. Ten assistants and ten services meant a hundred integrations. In practice, that meant very few existed.
What MCP is
Section titled “What MCP is”The Model Context Protocol is an open standard for how AI assistants talk to external tools and data sources. It was introduced by Anthropic and has been adopted well beyond it.
The useful analogy is USB. Before it, every device had its own connector. USB defined one standard, so any device works with any port.
MCP does that for AI assistants. A service exposes an MCP server describing what it can do. Any MCP-capable assistant — Claude, ChatGPT, Cursor and others — can connect and use it. One integration each side, rather than one per pair.
What an MCP server actually offers
Section titled “What an MCP server actually offers”An MCP server publishes a list of tools. A tool is a named capability with a description and defined inputs.
A portfolio MCP server might offer tools like:
query_holdings— list what the user ownsquery_performance_summary— return gains and lossesquery_user_fire_status— return progress towards financial independenceadd_transaction— record a buy or sell
When you ask your assistant a question, it reads the available tools, decides which are relevant, calls them, and uses the results to answer.
Crucially, the assistant is not guessing. It is reading live data through a defined interface.
Local versus remote
Section titled “Local versus remote”Two kinds exist.
Local servers run on your own machine, typically to reach local files or applications. You install and run them yourself.
Remote servers are hosted by the service and reached over the internet. Nothing to install — you point your assistant at a URL and sign in.
Agni Folio’s MCP server is remote, at https://agnifolio.com/mcp.
The security model
Section titled “The security model”This is the part worth understanding properly, because it is what makes connecting an AI to your finances reasonable rather than reckless.
You authenticate, not the AI. Connecting sends you through a normal sign-in and consent screen. You are approving access; the assistant never sees your password.
Access is scoped. Permissions are granular. Agni Folio uses two — reading your data, and writing to it — and you choose at consent time. Read-only is a genuine option.
Writes require confirmation. On a well-built server, anything that changes your data stops and asks. In Agni Folio, mutating tools produce a confirmation card inside the app that you approve before anything happens. The AI proposes; you decide.
It is revocable. You can withdraw access at any time from your account settings, and the connection stops working immediately.
It is logged. Every call is recorded, so you can see what was accessed and when.
Those five properties are the checklist. Apply them to any MCP server you are considering, not just this one.
What it feels like in use
Section titled “What it feels like in use”Once connected, you stop switching contexts. In the same chat window where you draft emails, you can ask:
What is my technology exposure across all my accounts?
Am I still on track for financial independence at 55?
What did I earn in dividends last quarter?
And get answers from your real portfolio.
Is this for you?
Section titled “Is this for you?”Honestly: it is optional. Everything in Modules 1 to 4 works without it. If asking questions inside the app suits you, you have already got the value.
Connecting your own AI is worth it if you already live in Claude or ChatGPT all day and want your finances reachable from there, or if you want to combine portfolio data with other reasoning in one conversation.
Practice step
Section titled “Practice step”Check whether the assistant you use supports MCP connections. Claude Desktop, ChatGPT (on paid tiers, via developer mode), Cursor, Cline and Continue all do, though the menu names differ.
If yours does, the next two lessons walk through connecting it.
Next: Connect Claude to your portfolio
Full technical details, including the complete tool list, live in the Connect Your AI documentation.