Fermi

Takes a natural language question like "Will this feature ship by Q2?" and returns calibrated probability distributions instead of point estimates. Works over both REST and MCP, so you can integrate forecasting into Claude conversations or your own applications. Useful when you need quantified uncertainty rather than binary predictions, like estimating project timelines, evaluating risks, or making decisions under uncertainty. The SSE transport means you can stream probability updates as the model refines its analysis. Reach for this when you want Claude to think probabilistically and give you distributions you can actually use in downstream calculations or decision frameworks.