"Did anyone in Congress trade this before the guidance cut?" is not a quote lookup. Neither is "which of my holdings added export-licence language this year?" Both live in filing text and disclosure records, and most market-data APIs do not carry that at all, so the agent guesses, which is worse than it saying no.

So I scored eight MCP servers for stock market data properly, on the criteria that decide whether an agent can answer or has to bluff.

Disclosure up front: I build one of these (Equibles) and it comes first. The criteria are below so you can disagree with them. They are weighted toward research rather than execution, which is where my own bias sits. Coverage is from public docs as of August 2026.

The six criteria

Each scored 0 to 5.

Primary source. Can the agent reach filing text, or only numbers someone else extracted?

Disclosure data. Congress trades, insider transactions, 13F, short interest.

Breadth per connection. How much one server answers before you have to add a second.

Cost of the first useful query. What a free tier lets an agent actually do, not the raw call count.

Ergonomics. Remote hosting, clean auth, official maintenance, tool descriptions written for a model rather than a developer.

Market-data depth. Latency, tick and order-book access, live chains.

That last one is the one I lose. I included it because a rubric that only measures your own strengths measures nothing.

The ranking

The spread comes almost entirely from the first two columns. Seven of eight score zero or near zero on primary source and disclosure. That is not a criticism. They are market-data businesses, and nobody builds a filings corpus by accident.

What each one actually carries

The scores above are my weighting. This part is just fact.

The filing-text column is the one that surprised me. Plenty of these will hand you a link to a 10-K. Almost none let the agent search inside the document and quote a line back with a position. If your agent needs to justify an answer rather than assert it, that column is the whole game.

Equibles MCP server

SEC filing text the agent can search inside, semantically or literally. Speaker-tagged earnings-call transcripts. Congressional trades, insider transactions, 13F and short interest. XBRL fundamentals with extracted KPIs, guidance and buyback programmes. Screening, options chains with greeks, and live US equity quotes. 108 tools behind one remote connection over OAuth.

Paste the URL into Claude or ChatGPT, no key to mint, and the same account answers plain REST if you would rather not speak MCP.

claude mcp add --transport http equibles https://mcp.equibles.com/mcp

Free tier is 100 calls a day with no dataset held back: the cap is the call count, not the catalogue, so an agent can try the filing search before anyone pays. Plus is \(9.99/mo, Pro is \)19.99/mo.

Financial Modeling Prep MCP server

Income statements, balance sheets, cash flow, ratios, valuation multiples, plus transcripts and 13F. Around $19/mo, and the biggest free tier here at 250 requests a day.

You get the extracted value but never the document, so an agent cannot audit a figure back to the filing or read the paragraph explaining a move. No chains.

Alpaca MCP server

Stocks, ETFs, crypto and options, plus brokerage. It is the only one here that can place an order and manage positions rather than just read, and paper trading runs against real data.

Free real-time IEX at 200 requests a minute, $99/mo for full SIP plus OPRA. No filings, no disclosure data.

Polygon MCP server (now Massive.com)

Trades, quotes, aggregates and live option chains across US equities, options and FX, at tick resolution, from around $29/mo. Chains are included rather than tiered away, which is rarer than it should be.

Architecturally it is the most interesting server here: three composable tools rather than one per endpoint, with SQL over results in an in-memory database. Purely market data beyond that.

Unusual Whales MCP server

Options flow, dark pool prints, Greek exposure, volatility surfaces and congressional trading across 100+ endpoints. Flow has no equal on this list.

$50/mo is the floor and there is no free tier at all, so you cannot let an agent try it first.

Alpha Vantage MCP server

Equities, FX, crypto, 50+ technical indicators, and options with all five greeks plus open-interest history back to 2008. Widest asset-class spread here.

Free tier is 25 calls a day, which is a demo rather than a trial. Real-time equities \(99.99, real-time options \)199.99.

Databento MCP server

Trades, OHLCV, full order-book depth, historical and live, with OPRA across all 17 US options exchanges. Deepest raw data on the list.

\(125 signup credit, usage-based history, OPRA live from \)199/mo. Worth knowing for this list specifically: its MCP servers are community projects rather than official, so tool signatures and support are not vendor-backed.

EODHD MCP server

60+ exchanges, 150,000+ tickers, 30 years of history across equities, ETFs, FX, crypto and macro. The one to reach for if the universe is not US-only.

US options are an end-of-day marketplace add-on, and the free plan is 20 calls a day capped to a year of history.

Where mine actually loses

3 out of 5 on market-data depth, and that is the honest score.

On the self-serve plans quotes ride IEX rather than full SIP, chains lag 15 minutes and carry greeks but no bid/ask, and there is no tick data or order book at any tier. Coverage is US equities only, so no forex, no crypto, no international listings.

If your agent needs microsecond fills or a full order book, Databento or Polygon is the right answer and I would rather say so here than waste your afternoon.

The test to run yourself

Connect whichever server you are evaluating and ask it this:

Did management raise full-year guidance on the most recent earnings call, and quote the exact words they used.

A server holding the transcript finds the passage and quotes it. A server without it either says it cannot answer, which is the honest failure, or reasons from prices and hands you something that sounds right and is not. That second outcome is the expensive one, and it is why the document layer matters more than another price feed.

Free tier at equibles.com/mcp is 100 calls a day with no card if you want to run that test today.