@withpica/mcp-server 2.90.0 → 2.91.1

package/CHANGELOG.md

@@ -9,6 +9,36 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 > matching entry here in the same commit. See the project's "npm publish
 > discipline" memory entry for the enforcement rationale.
 
+## [2.91.1] - 2026-07-08
+
+### Added
+
+- **`pica_tool_details` now returns an `invoke` pointer on every tool detail**
+  (single-tool and category paths). On the layered 3-tool surface the wrapped
+  tool is not in the caller's function list; models that fetched a schema and
+  found no matching callable concluded the tool was unavailable and gave up
+  (observed on ChatGPT). The `invoke` field spells out the execution path:
+  run via `pica_execute` with `{"tool": "<name>", "args": {...}}`.
+- **`pica_resolve_person` / `pica_resolve_work` are now natural-language
+  creator entries** ("find more information about this person / this song",
+  "enrich this…", "look this song up in the MLC"), with Creator Question
+  Atlas backing. Flipped from specialist after founder telemetry showed the
+  ask arriving in exactly those words and neither resolver surfacing.
+
+## [2.91.0] - 2026-07-07
+
+### Added
+
+- **The Oracle (Workstream 2) — count / rank / aggregate.** `pica_catalog_stats`
+  now accepts `entity` (works / recordings / releases / people) and `group_by`,
+  answering "how many have I produced", "who's my top collaborator", "genre
+  breakdown", "releases per year" in one call. No args returns the flat
+  multi-entity overview exactly as before (backward-compatible); `entity` alone
+  returns that entity's total (now including releases); `entity` + `group_by`
+  returns ranked buckets. Backed by org-scoped `GROUP BY` and a corrected
+  distinct-work collaborator ranking. No new tool — the existing counts tool is
+  enriched, and it is now surfaced as a natural-language creator-entry.
+
 ## [2.90.0] - 2026-07-07
 
 ### Changed