# Options Scanner — automated multi-strategy trade builder Also packaged as an **MCP server** (7 tools: scan, build, position/roll checks) for Claude Desktop, Claude Code and Hermes — see [MCP_SERVER.md](MCP_SERVER.md) for setup and an honest evaluation of the engine. [![Built with ML Intern](https://huggingface.co/datasets/huggingface/badges/resolve/main/built-with-ml-intern-sm.svg)](https://hf.co/chat?mode=ml-intern) Automated options scanner: pulls option chains + stock quotes from free online sources (Nasdaq/NYSE listed underlyings via Yahoo Finance), builds multi-leg strategies with a defined-risk-first design, gates every candidate through a risk engine, and outputs ranked trades as JSON/CSV for the browser dashboard. ## Data sources | Provider | Latency | Setup | Adapter | |---|---|---|---| | Alpaca (IEX) | real-time | `ALPACA_KEY_ID` + `ALPACA_SECRET_KEY` env vars (free) | `AlpacaProvider` | | Tradier | real-time | `TRADIER_API_KEY` env var (free with brokerage account) | `TradierProvider` | | Polygon.io | real-time | `POLYGON_API_KEY` env var (options plan) | `PolygonProvider` | | Yahoo Finance | ~15 min delayed | none (`pip install yfinance`) | `YFinanceProvider` (default) | The scanner auto-selects the best provider present: **Tradier > Alpaca > Polygon > yfinance**, or force one with `--provider`. All providers share one normalized `Chain` model, so switching feeds changes nothing else. Alpaca's free IEX feed has no open interest and thin one-sided quotes on some contracts; the engine treats one-sided quotes as wide markets (penalized in scoring, never silent) and drops quotes with impossible implied vols. > Honest note: "real-time" free options data does not exist anywhere. Free > feeds are delayed ~15 min. Plug in a Tradier key (free tier with a broker > account) for real-time chains. ## Strategies built Defined-risk by default — every short leg is paired inside a spread: - Verticals: bull/bear credit spreads and bull/bear debit spreads - Calendars (call & put, same strike) - Diagonals (long theta, offset strikes across expirations) - Long butterflies and Iron butterflies - Iron condors - Call ratio backspreads (1x2) - Jade lizards (credit > call-spread width where available) - Double diagonals (the defined-risk "short strangle") - Short straddles/strangles: rejected by default (undefined risk); pass `--allow-undefined-risk` to admit them with a loud warning — the defined-risk equivalents (iron fly / double diagonal) are built instead ## Risk engine - Payoff-grid max profit / max loss / breakevens at the shortest leg's expiry; back-month legs valued with Black-Scholes at that date (honest calendar and diagonal risk). Max loss is floored at the debit actually paid. - Gates: per-leg open interest, bid-ask spread vs mid, positive bids, defined risk, buying-power effect ≤ max % of account, earnings dates inside the trade window. - POP from lognormal distance of short strikes; composite score = expectancy (45%) + POP (25%) + credit efficiency (20%) + liquidity (10%). ## Rolling mechanics `rolling.RollManager.check(position, chain, rules)` marks open positions and returns concrete actions: close at profit target (default 50% of credit), stop at 2x credit, and rolls — out, out-and-down (tested put), out-and-up (tested call) — when DTE hits the management window or a short strike's delta breaches the tested threshold. ## Quick start ```bash pip install yfinance # auto universe (40 large caps + high-IV names + ETFs), free delayed data python -m options_scanner.scan --output results.json --csv results.csv # your watchlist, real-time via Alpaca export ALPACA_KEY_ID=... ALPACA_SECRET_KEY=... python -m options_scanner.scan --symbols SPY,QQQ,NVDA --provider alpaca ``` Tune per-trade risk with `--account-size` and `--risk-pct`. ## Output `results.json` carries `top_candidates` (ranked by composite score, best 3 per symbol) with full leg lists, breakevens, BPE in dollars, POP, and the reasons each trade passed/failed gates; `errors` lists tickers that failed to load. The JSON is what the browser dashboard reads. ## Disclaimer Educational tooling, not financial advice. Option pricing uses delayed data unless a real-time key is configured; verify every structure in your broker before trading.