@withpica/mcp-server 2.83.1 → 2.84.0

package/.claude/settings.local.json

@@ -0,0 +1,5 @@
+{
+  "disabledMcpjsonServers": [
+    "pica-directory"
+  ]
+}

package/CHANGELOG.md

@@ -11,16 +11,64 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
-## [2.83.1] - 2026-06-26
+## [2.84.0] - 2026-06-30
+
+### Added
+
+- **`pica_audio_update`** (#1379) — set `file_type` / `stem_label` /
+  `version_label` / `title` on an audio file after upload, wrapping the
+  org-validated `PATCH /audio-files/[id]` route.
+
+### Changed
+
+- **`pica_audio_query`** (#1378) — unassigned audio can now be filtered by
+  filename/title, making discover-and-assign workflows precise on large
+  upload sets.
+
+## [2.83.2] - 2026-06-27
+
+### Added
+
+- **`pica_statement_ingest_document`** (`write:finances`, `approvalMode:"human"`) —
+  agent-driven royalty/streaming statement ingestion for non-CSV statements
+  (PDF/image/xlsx). The agent reads the document, extracts the line items, and
+  calls this tool; for a delegated (OAuth/grant) caller it does NOT write — it
+  raises a human-approval request, and only the org owner's approval replays it
+  into `royalty_payments`, reusing the existing CSV parser downstream (identity
+  resolution, currency normalisation, fingerprint dedup). CSV statements stay on
+  the deterministic importer (PICA #1332).
+- **`pica_statements_pending`** (`read:finances`) — lists uploaded
+  `royalty_statement` documents the user hasn't imported yet (lean summaries, no
+  financial figures), so the agent can discover them; routes to
+  `pica_statement_ingest_document` (PICA #1332).
+- **`pica_audio_assign`** (`write:files`) — attach an uploaded audio file to a
+  work and/or recording after the fact; pairs with the new
+  `pica_audio_query unassigned:true` filter so an agent can discover orphan
+  audio and give it a home. Wraps the existing org-validated
+  `PATCH /api/admin/audio-files/[id]`; no approval gate (low-risk, reversible)
+  (PICA #1337).
 
 ### Changed
 
+- **Society-agnostic reads (ADR-278).** Bumped `@withpica/mcp-sdk` to `^2.0.0`,
+  whose read types replace the `prs`/`ppl`/`mlc` registration scalars with a
+  single `society_registrations[]` array, and updated the `pica_works_query`
+  tool descriptions so they no longer privilege PRS/PPL over other collection
+  societies. No tool was added or removed and no handler logic changed; this is
+  a republish so external stdio consumers receive the BREAKING SDK read shapes
+  already live on the hosted `/api/mcp`. Consumers reading `prs`/`ppl`/`mlc`
+  fields must migrate to `society_registrations[]`.
+- **`pica_audio_query`** — new `unassigned` boolean filter returns audio with no
+  work and no recording (PICA #1337).
 - **`pica_documents_read`** — description now tells the agent to STOP retrying
   when `parsing_status` is `failed` and read the new `parse_error` field for the
   reason (e.g. a legacy `.xls` that must be re-saved as `.xlsx`), instead of
   polling a permanently-failing parse. Pairs with the server-side fix that makes
   the read return an honest `failed` + `parse_error` rather than looping in
   `processing` forever (ops_issue 8e197fe5). No tool/schema/SDK change.
+- **`pica_documents_read`** — also points the agent at
+  `pica_statement_ingest_document` when the document is a royalty/streaming
+  statement (PICA #1332).
 
 ## [2.83.0] - 2026-06-25