senado-br-mcp 3.18.2 → 3.18.3

package/README.md

@@ -572,7 +572,7 @@ In addition to the `provenance` envelope, `structuredContent` carries a top-leve
 
 **Field-level granularity.** Most tools are single-source, so one envelope suffices. The few that merge reads in one response fill `field_sources` — a list of `{ fields, source_url, dataset_id, data_vintage, retrieved_at, served_from_cache }` attributing specific output fields to their real origin, and the block's `retrieved_at` is the oldest of them. `served_from_cache` is `true`/`false` for reads through the server cache and `null` where the notion does not apply (the bundled org-chart snapshot, the e-Cidadania corpus in D1). The merging tools (since 3.18.2): `senado_obter_materia` `secao=detalhe` (`/processo/{id}` + `ementa` from `/processo` + `relator` from `/processo/relatoria`), `senado_obter_senador` and the Deep Research `fetch` of a senator (biography + `mandatos`/`emExercicio` from `/senador/{codigo}/mandatos`), `senado_reunioes_comissao` over a window that crosses the year (one read per year), `senado_servidores` with `subordinadasA` (payroll + org-chart snapshot), `senado_suprimento_fundos` statistics on `atos-concessao` (acts + beneficiary names from the supridos registry), `senado_ecidadania_obter_evento` (live detail + `comentarios` from the corpus) and `senado_ecidadania_sugerir_tema_enquete` (consultas and ideias corpora). **Only a read that becomes a field of the answer is a sub-source**: a read that only resolves or checks an identifier (a committee acronym → its code, the parent check in `senado_contratacao_detalhe`) is left out, otherwise a cached lookup table would make the whole answer claim to be as old as it. `attribution` always lists every distinct underlying `source_url`.
 
-`retrieved_at` fidelity is provided by the cache layer (`cachedFetchWithMeta`), which persists the fetch timestamp alongside the value, so it reflects the real upstream extraction even on a cache hit. Two exceptions report an honest live timestamp instead: the e-Cidadania **list** tools (read from D1) use the corpus's `lastScrapedAt` — the true age of the stored data — while e-Cidadania **detail** tools, scraped live, use the fetch time and a level-3 canonical item URL. The only path that falls back to the build-time default is the in-code static reference catalog (`senado_tabelas_referencia` `tipos-materia`), which has no upstream extraction instant.
+`retrieved_at` fidelity is provided by the cache layer (`cachedFetchWithMeta`), which persists the fetch timestamp alongside the value, so it reflects the real upstream extraction even on a cache hit. Two exceptions report an honest live timestamp instead: the e-Cidadania **list** tools (read from D1) use the corpus's `lastScrapedAt` — the true age of the stored data, or the live scrape instant when the corpus is empty or stale and the list falls back to the portal (since 3.18.3; before, the stale corpus's instant leaked into the live answer) — while e-Cidadania **detail** tools, scraped live, use the fetch time and a level-3 canonical item URL. The only path that falls back to the build-time default is the in-code static reference catalog (`senado_tabelas_referencia` `tipos-materia`), which has no upstream extraction instant.
 
 Coverage is **universal**: all 69 tools carry the envelope — the 67 `senado_*` tools via `resultWithProvenance(` (verify with `grep -c 'resultWithProvenance(' src/tools/*.ts`) and the two Deep Research tools via `provenanceExtras` (same block, in `structuredContent`/`_meta`, since their text channel is the contract's JSON). The **⊕** marks in the inventory below denote the original pilot tools (votes, bills, processes); the envelope now extends to every tool, so the marks are historical.
 

package/dist/scraper/store.js

@@ -99,7 +99,11 @@ export async function resolveList(db, entidade, staleMaxMin, liveScrape, now = n
             items: live,
             meta: {
                 fonte: "ao-vivo",
-                lastScrapedAt,
+                // The items were scraped NOW, so this is their age — not the stale D1's `lastScrapedAt`
+                // read above (until 3.18.2 it leaked here, and the tools' provenance said the live data
+                // was as old as the corpus, or carried no instant at all). An empty live list extracted
+                // nothing (consultas_votos' live path is a no-op `[]`), so it claims no instant.
+                lastScrapedAt: live.length > 0 ? new Date().toISOString() : null,
                 possivelDesatualizacao: false,
                 motivo: !db ? "d1-indisponivel" : d1Items.length === 0 ? "d1-vazio" : "d1-desatualizado",
             },

package/dist/version.js

@@ -1,3 +1,3 @@
 /** Single source of truth for the server version (kept in sync with package.json). */
-export const VERSION = "3.18.2";
+export const VERSION = "3.18.3";
 export const USER_AGENT = `senado-br-mcp/${VERSION} (+https://senado.sidneybissoli.com; sbissoli76@gmail.com)`;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "senado-br-mcp",
-  "version": "3.18.2",
+  "version": "3.18.3",
   "description": "MCP server for Brazilian Senate open data — runs hosted on Cloudflare Workers or locally via npx (stdio)",
   "keywords": [
     "mcp",
@@ -83,7 +83,7 @@
     "@sbissoli/mcp-evals": "^0.2.1",
     "@sbissoli/mcp-surface": "^0.6.0",
     "@types/node": "^26.5.1",
-    "agents": "^0.26.0",
+    "agents": "^0.28.0",
     "exceljs": "^4.4.0",
     "tsx": "^4.23.13",
     "typescript": "^7.0.2",