What is an MCP server for investing? A plain-English guide

An MCP server for investing is a service that gives an AI assistant, like Claude or ChatGPT, a set of tools for market data and, optionally, your own portfolio. The assistant calls those tools when your question needs them, so its answer is based on current news, earnings dates and holdings rather than on whatever it remembers from training. MCP is the open standard that makes this work the same way across different assistants.

If you've wondered why your chatbot doesn't know what happened to your stock yesterday, this is the missing piece.

What is MCP, in one paragraph?

The Model Context Protocol is an open standard for connecting AI assistants to outside tools and data. Before MCP, every AI app had its own plug-in format. Now a single server can work with many assistants. The server describes its tools (a name, what the tool does, what inputs it takes), and the assistant decides when to call them. Think of it as a universal adapter between a chatbot and the rest of the world.

Why investing needs it

Language models are trained on a snapshot of text. That makes them good at explaining concepts like P/E ratios or what guidance means. It makes them bad at anything time-sensitive:

• They don't know today's news on a company.
• They don't know when the next earnings date is.
• They don't know what you own.

Investing questions are almost all time-sensitive. "Should I be paying attention to this headline?" depends on the headline, the date and your position. An MCP server supplies the facts. The model supplies the explanation.

What a stock market MCP server typically offers

Tools vary between servers. Broadly, you'll see some mix of:

As an example, Bywell's MCP server exposes your own data: get_holdings, get_allocation, watchlist tools (get, add, remove), your theses and notes, savings goals, contribution room, earnings reminder tools (create, list, delete), get_lesson for short explainers, and get_account for your plan. Live prices and news come from the assistant's own web search. There's no trading tool, on purpose.

How it works when you ask a question

Here's what happens behind the scenes when you ask "Which of my holdings report earnings in the next two weeks?":

1. The assistant reads your question and sees it has a tool called get_holdings.
2. It calls that tool and gets back your positions.
3. It searches the web for each company's next earnings date.
4. The assistant turns that into a readable answer with sources, and might offer to set reminders.

You don't see most of this. You just see an answer that's grounded in real data. Many assistants will show you which tool they called if you expand the response, which is a good habit to check.

Hosted vs self-hosted MCP servers

There are two ways to use an investing MCP server.

Hosted. Someone else runs the server. You add its address to your assistant as a connector and sign in. This is how most people will use MCP, and it doesn't need any code. Bywell works this way.

Self-hosted. You run the server yourself, often from an open-source project. You get full control, but you're responsible for API keys, data licences, uptime and security. This suits developers and tinkerers.

If you just want answers about your own portfolio in Claude or ChatGPT, hosted is the practical choice. Our guide to connecting your brokerage to ChatGPT or Claude walks through the steps.

What to look for in an investing MCP server

Before you connect one, ask these questions.

• Read-only or trading? Read the tool list. If there's an order or transfer tool, the assistant could in principle call it. Read-only servers remove that risk entirely.
• Where does the data come from? News and fundamentals should come from a named data provider, and news should carry a source and a date.
• How does it reach your brokerage? A reputable aggregator is safer than a server that asks for your brokerage password directly.
• What coverage does it have? Some servers only cover US stocks. If you hold TSX-listed shares, check for Canadian coverage.
• How does it tell the assistant to behave? MCP servers can send instructions to the assistant when it connects. A responsible investing server tells the assistant to explain rather than recommend.

That last point matters more than it sounds. Bywell's server instructions ask the assistant to explain what news could mean, give the bull and bear reading, cite the source and date, offer a short lesson when a concept comes up, and never tell you to buy, sell or hold a specific security or suggest position sizes.

What MCP doesn't fix

MCP gives the assistant better inputs. It doesn't make the assistant infallible. It can still misread a number or overstate a headline. A few habits help:

• Ask where a claim came from and check the date.
• Ask for the bear case as well as the bull case.
• For anything important, read the primary source, such as the company's filing on SEC EDGAR or SEDAR+ in Canada.

This is general information, not investment advice.

MCP connector vs a finance chatbot

You might wonder how this differs from an investing chatbot built into a finance app. The main difference is where the conversation happens.

• A built-in chatbot lives inside one app. It only knows that app's data, and you have to go there to use it.
• An MCP connector brings the data to the assistant you already use. The same connector can work in Claude, ChatGPT, Codex or Gemini, and it sits alongside your other connectors, like your calendar or notes.

That second point is easy to underrate. If you already do your thinking in one assistant, adding your portfolio there is less friction than learning another app. It also means you can mix sources in one conversation, for example asking about a company's earnings and then drafting your own notes on it.

The trade-off is that the quality of the answers depends partly on the assistant you choose. The connector supplies the same facts to each, but they explain them in their own way.

Getting started

If you want to try an investing MCP server without writing code, Bywell connects your brokerage read-only and works in Claude, ChatGPT, Codex, Gemini and other MCP clients. It's coming soon to iPhone and Android: join the launch list to hear the day it's out.