AI & Agents
AI agents design, execute, and iterate on cross-asset derivatives backtests through MCP — mixing options, futures, perpetuals, and spot in a single conversation. Theoretical and listed pricing modes reveal the gap between model edge and executable edge.
Traditional backtesting workflows are slow. Define a strategy in code, configure data pipelines, run the backtest, parse the output, adjust parameters, rebuild, re-run. Each iteration takes minutes to hours depending on complexity. For systematic teams evaluating dozens of parameter combinations across multiple assets, the iteration cost is the bottleneck.
MCP changes this by putting the backtester inside the conversation. An AI agent can define a strategy, execute it, inspect the results, adjust parameters, and re-run — all within a single session. The bi-directional protocol means the agent refines its approach based on what the data reveals, not what the developer anticipated upfront.
The agent can compose strategies that mix instrument types in a single backtest. Options, futures, perpetuals, and spot — each component with its own entry, exit, side, quantity, and underlying asset. A single backtest can hold BTC options alongside ETH perpetuals alongside SOL spot, with consistent P&L attribution across all legs.
An agent starts broad — "backtest a 30-day ATM straddle on BTC with hourly delta hedging using perpetuals." It inspects the results: Sharpe ratio, max drawdown, win rate, funding costs. Then it narrows — "same strategy but entry only when the 7d/30d term structure spread exceeds 3 points." Then it extends — "add a short ETH 25-delta strangle as a cross-asset hedge." Each iteration builds on the last without rebuilding the pipeline.
Prices options using SVI-calibrated implied volatility surfaces, spot prices, and forward curves. Shows what an option should have been worth at each timestamp based on the calibrated surface. Useful for strategies that trade theoretical edge — the model says the option is cheap, so buy it.
Prices against actual exchange bid-ask quotes. Snaps to the nearest available listed instrument matching the requested strike and tenor. Shows what you would have actually paid. If listed quotes are unavailable, falls back to theoretical with a warning so the backtest never silently gaps.
The agent can run the same strategy in both modes and compare. Theoretical P&L minus listed P&L equals the execution cost of the strategy — bid-ask spread, slippage, and the gap between where the model prices and where the market trades. This is the difference between a strategy that works on paper and one that works in production.
Options strategies can be delta-hedged with spot or perpetuals. The agent specifies the hedge instrument and rebalance frequency — one-shot at entry, hourly, or any interval. The backtest output separates strategy P&L, hedge P&L, and funding P&L for clean attribution.
Perpetual hedges accrue funding costs. The backtester tracks funding P&L separately, so the agent can evaluate whether the carry cost of the hedge erodes the strategy’s edge. An agent can compare spot-hedged versus perp-hedged versions of the same strategy to quantify the funding drag.
Each backtest returns per-trade records with entry and exit prices, Greeks, P&L, and funding. Per-instrument summaries show contribution by leg. The overall summary includes net P&L, win rate, Sharpe ratio, and maximum drawdown. Warnings flag any data fallbacks — if a listed price was substituted with theoretical, the agent knows and can factor that into its assessment.
The agent doesn’t just receive results — it interprets them and proposes refinements. Sharpe too low? Tighten the entry filter. Max drawdown too deep? Add a stop-loss or reduce position size. Funding drag eating the edge? Switch the hedge from perpetual to spot. Each cycle takes seconds, not hours.
Historical data spans BTC, ETH, SOL, XRP, HYPE, ADA, SUI, and additional assets. Both composite and exchange-specific volatility surfaces are available for theoretical pricing. Listed pricing uses actual exchange quotes from Deribit, Bybit, OKX, GateIO, and additional venues. Multi-year historical depth for strategy development across multiple market regimes.
