Odysseus - AI-Powered Strategy Research

Odysseus is a local MCP server for autonomous trading strategy discovery. Instead of embedding its own AI model, Odysseus connects to any compatible AI agent and gives it a disciplined quantitative research environment. The agent can analyze real Alpaca market data, propose trading hypotheses, formalize them as structured strategy specifications, generate deterministic C# StockSharp strategies, compile them, run realistic backtests, optimize parameters, and iterate on new ideas. Odysseus freezes and hashes every dataset so results remain reproducible. Backtests include trading costs, spread impact, drawdown, exposure, trade statistics, concentration and detailed breakdowns by month, session, holding time and direction. To reduce overfitting, research is separated into development, validation and hidden closed data. The agent can iterate on the visible data, while the final closed slice remains unavailable until the end and is measured only as a final check. Walk-forward evaluation and stressed transaction costs are also included. Once a candidate has been fully researched, the same compiled strategy can be deployed to an Alpaca paper account for forward observation. Odysseus does not decide whether a strategy is “good.” It provides reproducible evidence, while the AI agent decides which hypotheses to improve, reject or continue exploring.