congress-whales 🐋 — the "INSANE Claude Trading Hack," but actually free

This is a working rebuild of the setup in that Instagram reel: connect Claude to live "smart money" market data and ask it questions in plain English, e.g.

"Pull every stock a member of Congress bought this week. Rank them by how many members bought each."

"Cross-check those tickers with the latest 13F institutional filings. Show me where Congress and the big funds are buying the same names."

The reel used the Unusual Whales MCP server. That server is free software, but it wraps the Unusual Whales API, which needs a paid subscription (their historical/flow tiers run into the hundreds of dollars a month). So the "FREE" in the reel is doing some quiet work.

This project needs no subscription and no API key at all. It pulls the exact same kind of data straight from the primary public sources the aggregators repackage:

Download

Grab the latest from Releases:

Both are self-contained — no Python, no setup. The app checks GitHub for newer versions on launch and shows an in-app Update button when one exists. (The Android app runs the same Python engine on-device via Chaquopy inside a WebView; the APK is built automatically by GitHub Actions on every release.)

Related MCP server: GovGreed MCP

Optional: richer news (Finnhub)

By default the app uses free Google News. For deeper, strictly date-ranged company news, get a free key at finnhub.io/register, copy config.example.json to config.json, and paste the key. config.json is git-ignored so it never ends up in the repo.

Two ways to use it

A) Desktop app — a standalone dashboard, no Claude needed. Two ways to launch:

• Congress Whales.exe on your Desktop — a single shareable file (~41 MB, no Python needed). Just double-click, or send it to a friend. (First launch is slow while it builds the first snapshot; Windows SmartScreen may warn because it's unsigned — click "More info → Run anyway". Needs an internet connection.)

• The Congress Whales shortcut / Congress Whales.bat — runs from source (for when you've edited the code). Rebuild the exe with build_exe.bat.

A compact, tabbed window opens, refreshed to the newest filings each time. Every table sorts — click any column header. Tabs:

• Overview — KPIs, an expandable "how to read this" explainer (auto-opens when net bias by transaction count disagrees with net $ flow by dollars), Sector trends with a measurable % change (newer half of the window vs older half — e.g. "Healthcare buying up +61%"), a weekly chart, a clear sector buys/sells breakdown, a 💡 biggest copyable winners teaser, and a Republicans vs Democrats comparison.

• Buys / Sells — leaderboards; each row shows sector, price, and return since the trade. Filter by sector or search.

• All trades — every disclosure, filterable by buy/sell, sector, chamber, and party, showing return since the trade and since it became public.

• Members — every trading member with party, sortable by any column (buys, sells, est. return, copyable return, volume…), filter by party, and search a specific member. Click one for their disclosed portfolio.

• Insights — 🏆 who to copy (members with the best return since their trades went public) and 💡 biggest copyable winners (individual disclosed buys, e.g. a PANW buy up ~67% since it was disclosed).

• Fund overlap — tickers Congress bought that marquee hedge funds also hold (13F).

Hover any ticker for a profile card (company, what it does, sector/industry, market-cap size, S&P 500 = ES / Nasdaq-100 = NQ). Click a ticker for a drawer with price, return-since-trade, the biggest price move near the trade date, the business summary, and news aligned to the trade (amber = before, teal = after).

Lookback windows run 7 days → 2 years. Opens instantly on the last snapshot, then updates in the background.

Copyable return — the "what could a regular person do" number

Members can file a trade up to ~45 days after making it, so you can't buy at their price. The app therefore also computes return since the filing became public ("copy return") — what you could realistically have made mirroring the trade after seeing the disclosure. That's what powers the Insights tab. Ideas, not advice.

Honest caveats (you asked for these)

• "Portfolio" = disclosed trading activity, not a verified holdings statement. Filings give dollar ranges, not share counts or cost basis — so net values are estimates.

• "Est. return" is unrealized mark-to-market since each disclosed buy.

• Realized profit and net worth are deliberately not shown — they can't be derived reliably from transaction reports (no lots, only ranges).

• Only Congress is included. House + Senate are the only U.S. bodies with a clean, transaction-level public feed (the STOCK Act). Executive-branch officials and federal judges file annual holdings disclosures (broad ranges, no clean transaction feed), so they can't be added here without a big drop in data quality.

• Party is matched by name to the congress-legislators roster (~98% hit rate; a rare member may show "?"). Index membership is a static ~2026 snapshot.

If the shortcut ever goes missing, recreate it by right-clicking create_shortcut.ps1 → Run with PowerShell. For troubleshooting, run Congress Whales (debug).bat (same app, but with a console that shows errors).

B) Inside Claude — the same data as MCP tools you can ask about in plain English.

What Claude can do once it's connected

Five tools show up in Claude:

• congress_most_bought — the leaderboard: stocks ranked by how many members of Congress bought them recently. (Reel prompt #1.)

• cross_check_congress_vs_funds — the overlap of what Congress is buying and what marquee hedge funds hold in their latest 13F. (Reel prompt #2.)

• congress_recent_trades — the raw individual disclosures (filter by chamber, buy/sell, ticker).

• institutional_holdings — which marquee funds hold a given ticker.

• list_marquee_funds — the institutions behind the cross-check (Berkshire, Citadel, Pershing Square, Renaissance, Point72, Tiger Global, Scion/Burry, and more).

Then you just talk to Claude:

"Use congress_most_bought for the last two weeks, then cross-check the top names against the funds and tell me which 3 have the strongest combined conviction."

Setup

1. Install dependencies

pip install -r requirements.txt

(Only two packages: mcp and pypdf.)

2. Prove it works (no Claude needed)

python server.py --selftest

You should see a live congressional buy leaderboard, real 13F filing dates, and a Congress-vs-funds overlap table. If that prints RESULT: PASS, the data pipeline works.

3. Connect it to Claude

Claude Code — a ready-to-use .mcp.json is already in this folder. Open Claude Code with this directory as the working folder and it loads automatically. To register it globally instead:

claude mcp add congress-whales -- "C:\Users\flami\AppData\Local\Programs\Python\Python312\python.exe" "C:\Users\flami\Desktop\congress-whales-mcp\server.py"

Claude Desktop — copy the congress-whales block from claude_desktop_config.example.json into your Claude Desktop config (%APPDATA%\Claude\claude_desktop_config.json) and restart Claude Desktop.

Want the exact reel parity (both chambers, options flow, dark pool)?

Add the official Unusual Whales MCP alongside this one — get a key at unusualwhales.com/settings/api-dashboard, then follow their guide at unusualwhales.com/public-api/mcp. This free server and the paid one can coexist; Claude will use whichever tool fits the question.

Honest limitations (so you trust the output)

• Windows. Only actual buys/sells reported; congress_most_bought defaults to common stock (include_all_asset_types=True adds options/bonds).

• "This week" = filing date. Members can legally file a Periodic Transaction Report up to ~45 days after the trade, so a fresh filing may describe an older trade. The trade's own date is included in every result.

• 13F is quarterly and lagged (up to ~45 days). A holding in the latest 13F means "held as of quarter-end," not necessarily "bought this quarter." Treat the cross-check as conviction overlap, not a real-time signal.

• Cross-check matches by company name (13F reports CUSIP + issuer name, not tickers). The matcher is deliberately strict to avoid false positives, so it can occasionally miss an unusual name.

• Senate depends on efdsearch.senate.gov being reachable; if it rate-limits, the server degrades gracefully to House-only rather than failing.

• Paper (non-electronic) filings are skipped — they're scanned images without machine- readable trade data.

Not financial advice

This is a data tool. Members of Congress and 13F filers disclose after the fact. Do your own research.