I've been running congressional trading questions through Claude with a financial-data MCP server connected, and the tool calls come back fast. The hard part is reading them. Members of Congress file periodic transaction reports within 45 days of a trade, and those reports are vague in very specific ways. If the assistant (or you) ignores that, the answers sound precise and are wrong.
Here are the three places it goes wrong, with numbers from a run on November 9, 2026. The full six-prompt walkthrough is in How to Track Congress Stock Trades With ChatGPT or Claude.
1. Every amount is a range
A disclosure says "$1,000,001 to $5,000,000", not a dollar figure. Any tool that ranks "what Congress bought" has to collapse that band into one number, and the usual choice is the midpoint. That's what GetMarketWideCongressionalActivity does: over the 30 days of filings to November 9, Costco led net selling at about $558,000 from two members, and the largest net purchases were about $75,000 each of Datadog and Estée Lauder.
Those are midpoint estimates. Ask the assistant to say so whenever it adds ranges together, and check who sits behind each row: one representative, Pete Sessions, is among the sellers in all five of the top selling rows, so the "trend" is mostly one portfolio.
Net worth works the same way. GetMemberNetWorth reads annual reports where assets and liabilities are also ranges, so Nancy Pelosi's 2025 report gives a band from -$66.6 million to $351.2 million. The negative floor comes from subtracting the top of every liability range from the bottom of every asset range.
2. The owner and instrument columns carry the meaning
GetMemberTrades for Pelosi returns 25 trades in the 12 months to November 9: 18 purchases and 7 sales. All 25 are in her husband's name, and 9 are options (asset type OP) rather than stock (ST). A summary that says "Pelosi bought Intel" without those two columns loses most of what the filing says. Tell the assistant to keep Owner and Asset Type in every table it builds.
The PDF often says more than the structured fields. The House Clerk filing behind the household's Bloom Energy purchases (20035143) lists 10,000 shares plus 100 call options with a $100 strike expiring June 17, 2027 on July 24, then 5,000 shares and another 100 of the same calls on July 28.
3. The filing lag sets the price you could have paid
The Bloom trades happened on July 24 and 28 and were filed on August 21. GetStockPrices gives the closes: $184.89 and $166.84 on the trade dates, $201.45 on the filing date. Anyone copying the trade could only buy from August 21, 9.0% and 20.7% above the trade-date closes. The stock peaked at $295.78 on October 6 and closed at $173.24 on November 6, 14.0% below the filing-day close.
So the useful question is never "what did they buy?" but "what could I have bought, and when?" Asking for the close on the trade date, the filing date and today forces the assistant to show that gap.
One prompt that covers all three
Show me what members of Congress traded most in disclosures from the last 30 days. Then, for Nancy Pelosi, list her last year of trades with trade and filing dates, owner and instrument, pick her largest purchase, show its stock's closing price on the trade date, the filing date and today, and list any other member who traded the same stock this year.
The assistant chains four tools for that: the market-wide ranking, the member's trades, the stock's trades across Congress, and daily prices. Setup for Claude or ChatGPT takes a few minutes.
Source: Equibles
Top comments (0)