mcp-scraper 0.57.1 → 0.57.3

package/CHANGELOG.md

@@ -4,6 +4,20 @@ All notable changes to MCP Scraper are documented here. The format is based on [
 
 ## [Unreleased]
 
+## [0.57.3] - 2026-08-16
+
+### Fixed
+
+- Local Sourcebook exact-place matching no longer rejects a Google listing that appends a service tagline to the business name. `nameOverlap` divided the intersection by the larger token set, so "Meljestic Spa" against "Meljestic Spa: Laser hair removal and Skincare" scored 0.33 against a 0.6 threshold and could never publish — even though the submitted name is fully contained in the Google name and Maps linked the same domain. Identity now also accepts containment of the smaller token set (which must carry two distinctive tokens) or a Maps listing that links the submitted domain. Wrong domain, wrong state, missing address, and single-generic-token still reject.
+- `acquisition_error` is now returned as `acquisitionError` on the submission. It was written on every failure and never selected back, so no response explained why a listing failed.
+- `local_sourcebook_submission_status` and `local_sourcebook_refresh` now state what a failed status means, that coverage counters are only written on success (so zeroed counters do not mean the crawl never ran), and that a replayed refresh still re-runs acquisition.
+
+## [0.57.2] - 2026-08-16
+
+### Fixed
+
+- Blob storage no longer falls back to the local filesystem on a serverless deployment. `BLOB_READ_WRITE_TOKEN` was empty in production, so `getBlobStore()` returned a local store and every write through it — the scrape-vault fallback document and Instagram media downloads — was saved to an ephemeral path inside the function and handed back as a `file://` URL no request could read. Without a token a serverless deployment now returns a store that refuses to write and says why.
+
 ## [0.57.1] - 2026-08-16
 
 ### Fixed

package/README.md

@@ -90,7 +90,7 @@ Build the branded one-click bundle:
 npm run build:mcpb
 ```
 
-The generated bundle is written to `build/mcpb/mcp-scraper-<version>.mcpb` and copied to `public/downloads/` for the hosted download. The current public bundle is `https://mcpscraper.dev/downloads/mcp-scraper.mcpb` (`0.57.1`, SHA-256 `ea84195321a47e7550348be6f61ed3e6a293a679c20a6cd39e77cd37870bd636`). Install it by opening or dragging it into Claude Desktop. Claude displays the `MCP Scraper` install card, icon, and API-key configuration field from the bundle manifest.
+The generated bundle is written to `build/mcpb/mcp-scraper-<version>.mcpb` and copied to `public/downloads/` for the hosted download. The current public bundle is `https://mcpscraper.dev/downloads/mcp-scraper.mcpb` (`0.57.3`, SHA-256 `b1edddd21790e7e5f24bacad85d992d8e995199c5d7e1d314500ba0abaab117c`). Install it by opening or dragging it into Claude Desktop. Claude displays the `MCP Scraper` install card, icon, and API-key configuration field from the bundle manifest.
 
 The MCPB install exposes every tool — web-intelligence plus all `browser_*` tools — through the one `mcp-scraper` server.
 

package/dist/bin/api-server.cjs

@@ -29050,6 +29050,7 @@ function mapSubmission(row) {
     businessName: String(rowValue(row, "business_name")),
     websiteUrl: String(rowValue(row, "website_url")),
     status: String(rowValue(row, "status")),
+    acquisitionError: rowValue(row, "acquisition_error") == null ? null : String(rowValue(row, "acquisition_error")),
     draftRevision: Number(rowValue(row, "draft_revision")),
     publishedRevision: rowValue(row, "published_revision") == null ? null : Number(rowValue(row, "published_revision")),
     coverage: parseJson2(rowValue(row, "coverage_json")),
@@ -30915,6 +30916,15 @@ function nameOverlap(left, right) {
   const intersection = [...a].filter((token6) => b.has(token6)).length;
   return intersection / Math.max(a.size, b.size);
 }
+function nameContainment(left, right) {
+  const a = tokens(left);
+  const b = tokens(right);
+  if (!a.size || !b.size) return 0;
+  const smaller = a.size <= b.size ? a : b;
+  if (smaller.size < 2) return 0;
+  const intersection = [...a].filter((token6) => b.has(token6)).length;
+  return intersection / smaller.size;
+}
 function validateLocalSourcebookMapsIdentity(submission, maps) {
   const expectedHost = host(submission.websiteUrl);
   const mapsHost = host(maps.website);
@@ -30924,7 +30934,9 @@ function validateLocalSourcebookMapsIdentity(submission, maps) {
   if (mapsHost && expectedHost && mapsHost !== expectedHost && !mapsHost.endsWith(`.${expectedHost}`) && !expectedHost.endsWith(`.${mapsHost}`)) {
     throw new Error(`Exact-place match rejected: Google Maps linked ${mapsHost}, not the submitted domain ${expectedHost}.`);
   }
-  if (overlap < 0.6) {
+  const containment = nameContainment(submission.businessName, actualName);
+  const domainConfirmsIdentity = Boolean(mapsHost && expectedHost);
+  if (overlap < 0.6 && containment < 0.9 && !domainConfirmsIdentity) {
     throw new Error(`Exact-place match rejected: Google Maps returned \u201C${actualName},\u201D which does not sufficiently match \u201C${submission.businessName}.\u201D`);
   }
   if (addressState && addressState !== submission.state.toLowerCase()) {
@@ -37303,16 +37315,21 @@ function byteLength(data) {
 function localBaseDir2() {
   return process.env.MCP_SCRAPER_OUTPUT_DIR?.trim() || (0, import_node_path11.join)((0, import_node_os7.homedir)(), "Downloads", "mcp-scraper");
 }
+function runsWithoutLocalDisk(env = process.env) {
+  return Boolean(env.VERCEL || env.AWS_LAMBDA_FUNCTION_NAME);
+}
+function resolveBlobStore(env = process.env) {
+  const token6 = env.BLOB_READ_WRITE_TOKEN?.trim();
+  if (token6) return new VercelBlobStore(token6);
+  if (runsWithoutLocalDisk(env)) return new UnconfiguredBlobStore();
+  return new LocalBlobStore(localBaseDir2());
+}
 function getBlobStore() {
   if (cached) return cached;
-  if (process.env.BLOB_READ_WRITE_TOKEN) {
-    cached = new VercelBlobStore(process.env.BLOB_READ_WRITE_TOKEN);
-  } else {
-    cached = new LocalBlobStore(localBaseDir2());
-  }
+  cached = resolveBlobStore();
   return cached;
 }
-var import_node_fs7, import_node_os7, import_node_path11, LocalBlobStore, cached, VercelBlobStore;
+var import_node_fs7, import_node_os7, import_node_path11, LocalBlobStore, cached, UnconfiguredBlobStore, VercelBlobStore;
 var init_blob_store = __esm({
   "src/api/blob-store.ts"() {
     "use strict";
@@ -37342,6 +37359,15 @@ var init_blob_store = __esm({
       }
     };
     cached = null;
+    UnconfiguredBlobStore = class {
+      kind = "unconfigured";
+      async put() {
+        throw new Error("Blob storage is not configured on this deployment: set BLOB_READ_WRITE_TOKEN. Refusing to write to a local path that no request can read back.");
+      }
+      async get() {
+        return null;
+      }
+    };
     VercelBlobStore = class {
       constructor(token6) {
         this.token = token6;
@@ -44624,7 +44650,7 @@ var PACKAGE_VERSION;
 var init_version = __esm({
   "src/version.ts"() {
     "use strict";
-    PACKAGE_VERSION = "0.57.1";
+    PACKAGE_VERSION = "0.57.3";
   }
 });
 
@@ -50633,14 +50659,14 @@ function registerPaaExtractorMcpTools(server, executor, options = {}) {
   }, async (input) => executor.localSourcebookCapture(input));
   server.registerTool("local_sourcebook_submission_status", {
     title: "Local Sourcebook Submission Status",
-    description: "Read the authenticated caller\u2019s listing draft, enrichment coverage, immutable revision number, publication state, and exact live LocalSourcebook.com profile and reviews URLs.",
+    description: "Read the authenticated caller\u2019s listing draft, enrichment coverage, immutable revision number, publication state, and exact live LocalSourcebook.com profile and reviews URLs. When status is failed, acquisitionError states why in plain language \u2014 read it before retrying, because a refresh repeats the same acquisition and will fail the same way until the cause is fixed. Coverage counters are written only when acquisition succeeds, so a failed listing still shows the pre-run crawl block; zeroed counters do not mean the crawl never ran.",
     inputSchema: LocalSourcebookSubmissionStatusInputSchema,
     outputSchema: recordOutputSchema("local_sourcebook_submission_status", LocalSourcebookOutputSchema),
     annotations: localPlanningToolAnnotations("Local Sourcebook Submission Status")
   }, async (input) => executor.localSourcebookSubmissionStatus(input));
   server.registerTool("local_sourcebook_refresh", {
     title: "Refresh a Local Sourcebook Listing",
-    description: "Queue a new broad crawl and review/media acquisition pass for a listing owned by the authenticated MCP Scraper account. A refresh costs 2 Credits total, including acquisition; idempotent retries are not charged twice. The last published revision remains public until the refreshed evidence revision completes and auto-publishes.",
+    description: "Queue a new broad crawl and review/media acquisition pass for a listing owned by the authenticated MCP Scraper account. A refresh costs 2 Credits total, including acquisition; idempotent retries are not charged twice, and a replayed retry still re-runs the acquisition, so a failed listing is always recoverable. Read local_sourcebook_submission_status first: a refresh repeats the same acquisition, so fix whatever acquisitionError reports before spending another pass. The last published revision remains public until the refreshed evidence revision completes and auto-publishes.",
     inputSchema: LocalSourcebookRefreshInputSchema,
     outputSchema: recordOutputSchema("local_sourcebook_refresh", LocalSourcebookOutputSchema),
     annotations: liveWebToolAnnotations("Refresh a Local Sourcebook Listing")