ai.ironscout/ironscout MCP server

Caliber-level price distribution (median, p25, p75, min, max) over a rolling window. Purely descriptive statistics with methodology and sample counts; never a forecast or recommendation. Returns `error_v1` `NOT_FOUND` when no snapshot exists for the caliber/window. When `extremaSuppressedByPolicy` is `true`, `min`, `max` and both `segments.*.min` / `segments.*.max` are `null` BY POLICY rather than for lack of data: that caliber's published extremes are withheld pending revalidation. The median, quartiles, `p95`, segment medians and counts remain valid and complete — `p95` is a percentile over the whole window rather than a single observation, so it is not withheld. Do not infer or substitute a floor for those calibers, and do not treat `p95` as a stand-in for the withheld `max`. This is distinct from `dataStatus: "INSUFFICIENT_DATA"`, which means the sample was too thin to compute anything.

Fetch a single canonical product by its `canonicalProductId`, with current per-retailer offers. Descriptive only. Returns an `error_v1` body with `NOT_FOUND` if the id is unknown.

Enumerate every canonical caliber IronScout recognizes, each flagged with whether a current market snapshot exists and its sample count. Calibers without a snapshot are still listed (with `hasSnapshot: false`).

List currently eligible/visible retailers. Ineligible retailers are omitted entirely (fail-closed, ADR-005). Descriptive directory only.

Search canonical ammunition products and their current per-retailer offers. Results are descriptive only — prices, availability, $/round, observation time, and provenance, with no verdicts, scores, or recommendations. Use each offer's `outUrl` for outbound links (do not rewrite it). `sort` is a mechanical ordering, never a judgement.