Realized-vs-Implied Options Trading Agent

VolEdge is an autonomous agent that trades options vertical spreads on Alpaca's paper platform, executing entirely through Alpaca's MCP server rather than direct API calls. Alpaca's data tier doesn't return implied volatility or greeks, so we built the pricing engine ourselves : Black-Scholes, a Newton-Raphson solver to invert market prices into implied volatility, and closed-form greeks, in pure Python with no external pricing library. The agent's edge comes from comparing implied volatility against realized volatility from the last 30 days of price history. When implied sits meaningfully below what the underlying has actually been moving, buying premium is favored; when it sits above, selling premium is. Every candidate spread is repriced at both measures, scored by edge, probability of profit at breakeven, and how many multiples of round-trip slippage the edge covers because a model that ignores execution cost isn't measuring real edge. Every proposed trade passes through a hard risk gate before reaching the broker: position sizing scaled by conviction, a portfolio-wide dollar-delta cap so several positions can't quietly become one large directional bet, per-underlying concentration limits, and a rule against duplicating the same directional view across correlated names. The agent can and does decide no candidate is worth trading on a given day. It scans 11 liquid US names, ranks candidates, sizes them, and executes multi-leg limit orders via MCP tool calls, with every decision, rejection, and fill logged to a journal for full auditability. It runs on a schedule, unattended, through Alpaca's paper account. Built solo across the hackathon window, including three real bugs found and fixed live: a parsing failure that made the agent believe its own positions didn't exist, an unpinned dependency that silently broke the MCP server mid-run, and a partial-fill flatten that left naked legs each one a genuine lesson in what "autonomous" actually requires.