Blog › Guides
What a financial data MCP server is, and how to connect one to Claude
Ask an AI assistant for a company's margin and it will give you a number with great confidence. The problem is where it comes from. An MCP server fixes that: it gives the assistant a real data source it can query.
16 November 2026 · 6 minute read
The problem: AI has memories, not data
A language model learns from text up to a date and then stops learning. What it knows about a company is what it read back then, mixed together and undated. Ask it about recent results, or about a small company few people write about, and it will usually get it wrong or fill the gap with something plausible.
For investing, that is not good enough. You need the figure to be current, to know which date it is from, and to get the same answer if you ask twice.
What MCP is
The Model Context Protocol (MCP) is an open standard Anthropic published in November 2024 for connecting AI assistants to external data sources and tools. Claude, Claude Code, Cursor, VS Code and many other clients use it today.
The idea is simple. An MCP server offers a list of tools, each with a description: "search companies", "compare these metrics", "give me the history of this figure". When you ask a question, the assistant decides which tools it needs, calls them, gets the data back and answers with it. You do not need to know any of that: you just ask.
What a fundamental data MCP server does
Market Pi-Rates is one of them. Behind it sits a database of every listed company in the United States, Europe, Japan, Hong Kong and Australia, around 25,000, updated every morning. These are the tools the assistant sees:
-
search_companiesandget_company: find a company and see all its figures. -
screen: filter the whole universe with conditions, like a screener. -
compareandsector_stats: put several companies side by side, or see a sector's median. -
metric_historyandprice_history: how a figure has changed capture by capture, and the price week by week. -
upcoming_earnings: who reports results in the coming days. -
list_metricsanddata_coverage: which metrics exist and how fresh the data is.
On top of the figures companies publish, there are almost 70 ratios we calculate ourselves: margins, ROIC, cost of capital, multiples on enterprise value, DuPont breakdown and expected growth.
How to connect it
In Claude (web and desktop)
- Go to Settings → Connectors and click Add custom connector.
- Give it a name and paste this address:
https://themarketpirates.com/mcp - Click Connect and sign in with your Google account. That is it: no tokens to copy.
Claude Code
claude mcp add --transport http --scope user market-pirates https://themarketpirates.com/mcp
Cursor, VS Code and other clients
Any client that supports remote MCP servers over HTTP works: add a server with the address above. The connection guide has the steps for each one.
What to ask it
"Compare Inditex, H&M and Fast Retailing on revenue growth, operating margin and valuation."
"Find Japanese companies with no net debt, a ROE above 12% and cheaper than their sector median."
"Who in the semiconductor sector reports results this week?"
The assistant answers with the figures and the date of the capture they come from, so you always know what you are looking at.
What it does not do
It provides data, not recommendations. The investment decision is yours, and before you make it you should check any important figure against the company's own published accounts. And like any database, it can contain errors: if something looks off, ask the assistant for that figure's history, which is the fastest way to spot it.
The free plan sees all 25,000 companies with today's data. Connect it and ask your AI the first thing that comes to mind.
Try it free