Several MCP servers connect Claude and ChatGPT to stock market data. Most are API passthrough servers that forward each request to a data provider like Alpha Vantage or Polygon. Every question triggers one or more external API calls, subject to rate limits and API key requirements.
Shibui Finance takes a different approach. It connects Claude to a pre-loaded database containing 64 years of US equity data: 31M+ daily price records, quarterly and yearly financial statements, 56 pre-calculated technical indicators, and valuation metrics for 9,500+ companies. No API keys, no rate limits, no per-query costs. Claude queries the database directly, which means cross-market screening and multi-period analysis run in seconds rather than requiring thousands of individual API calls.
The MCP options for financial data
The main MCP servers for stock market data each target a different slice of the problem. Alpha Vantage MCP offers a free tier with 25 requests per day, covering real-time quotes and technical indicators. Financial Modeling Prep (FMP) MCP provides 250+ tools with deep fundamentals and direct SEC filing access. EODHD MCP covers 70+ global exchanges with 75 tools and OAuth authentication. Polygon MCP focuses on real-time tick-level data for US markets. Yahoo Finance MCP is free and unofficial, pulling from Yahoo's public endpoints.
These are all API passthrough servers. When Claude needs data, it calls the provider's API, waits for the response, and works with whatever comes back. This architecture works well for targeted, single-company lookups. It becomes a bottleneck when the question requires checking thousands of companies or joining multiple data types across long time periods.
API passthrough vs pre-loaded database
The core difference is where the data lives at query time. With an API passthrough server, each question triggers one or more external API calls. Ask for a company's financials and you get one call. Ask to compare 20 companies and you get 20 calls. Ask to screen the entire market and you hit rate limits before you get through 5% of the universe.
With a pre-loaded database, the data is already there. Claude writes a SQL query that runs against 31M+ rows in a single pass. There is no external call at query time, no rate limit to hit, no API key to configure. This distinction matters most for cross-market screening - the kind of question where you need to check a condition across every company in the database.
"Screen all US stocks for companies with revenue growth above 20% for 4 consecutive quarters and positive free cash flow"
This query checks roughly 9,500 companies in one pass. Claude writes a single query that examines quarterly financials across multiple periods, filters for the growth condition holding in each consecutive quarter, and joins the results with cash flow data. With an API passthrough server, you would need thousands of individual API calls - one per company, often more - hitting rate limits and taking minutes or hours to complete.
What a pre-loaded database makes possible
Some questions are straightforward with a database and effectively impossible with individual API calls. These are queries that require joining millions of rows across multiple tables or checking conditions across long time periods.
"What was the average price change 30 days after stocks with RSI below 25 had a P/E below 15?"
"Find tech companies where operating margin expanded every year for 5 years"
The first query joins daily technical indicators with valuation data to find co-occurring conditions, then measures forward price returns at a fixed offset. It touches millions of rows across three tables. The second checks a directional trend across five yearly periods for every technology company in the database. Both run in seconds against a pre-loaded database. With an API passthrough server, you would need to fetch data for each company individually, compute the joins and conditions client-side, and handle the rate limiting across thousands of calls.
When API passthrough makes more sense
A pre-loaded database is not the right tool for every job. If you need real-time or intraday price data, Alpha Vantage or Polygon are better options. Shibui updates daily after market close, so it cannot answer questions about what a stock is trading at right now.
If you need international markets, EODHD covers 70+ exchanges worldwide. Shibui covers US equities only - NYSE and NASDAQ. If you need options chain data, Polygon or Alpha Vantage have that; Shibui does not. If you need the very latest quarterly filing on the day it is published, FMP's direct SEC feed may surface it faster than Shibui's daily import cycle.
Shibui's limitations
Shibui covers US equities only (NYSE and NASDAQ). Data is end-of-day, not real-time or intraday. There is no options, forex, or cryptocurrency data. The underlying data comes from tier-3 providers - not Bloomberg or Refinitiv. Each query returns a maximum of 200 rows. GICS sector classifications have some gaps, particularly for ETFs and closed-end funds. These are real constraints, and for some workflows they are disqualifying.