mcp-scraper 0.1.7 → 0.1.9

package/dist/bin/api-server.cjs

@@ -50,6 +50,109 @@ var init_harvest_timeout = __esm({
   }
 });
 
+// src/lib/browser-service-env.ts
+function browserServiceApiKey() {
+  const value = (process.env.BROWSER_SERVICE_API_KEY ?? process.env.KERNEL_API_KEY)?.trim();
+  return value || void 0;
+}
+function browserServiceProxyId() {
+  const value = (process.env.BROWSER_SERVICE_PROXY_ID ?? process.env.KERNEL_PROXY_ID)?.trim();
+  return value || void 0;
+}
+var init_browser_service_env = __esm({
+  "src/lib/browser-service-env.ts"() {
+    "use strict";
+  }
+});
+
+// src/errors.ts
+function sanitizeVendorName(message) {
+  return message.replace(/kernel\.sh\s+sessions?/gi, "sessions").replace(/kernel\.sh\s+session/gi, "this session").replace(/kernel\.sh/gi, "the service").replace(/kernel\s+sessions?/gi, "sessions").replace(/kernel\s+session/gi, "this session").replace(/\bkernel\b/gi, "the service").replace(/  +/g, " ").trim();
+}
+var RECAPTCHA_INSTRUCTIONS, CaptchaError, ExtractionError, RequestAbortedError;
+var init_errors = __esm({
+  "src/errors.ts"() {
+    "use strict";
+    RECAPTCHA_INSTRUCTIONS = "Google returned a CAPTCHA. Run with --headless=false to re-warm the browser profile, then retry.";
+    CaptchaError = class extends Error {
+      constructor(instructions) {
+        super(`CAPTCHA detected. ${instructions}`);
+        this.instructions = instructions;
+      }
+      instructions;
+      name = "CaptchaError";
+    };
+    ExtractionError = class extends Error {
+      constructor(message, cause) {
+        super(message);
+        this.cause = cause;
+      }
+      cause;
+      name = "ExtractionError";
+    };
+    RequestAbortedError = class extends Error {
+      name = "RequestAbortedError";
+      constructor(message = "Request aborted before harvest completed") {
+        super(message);
+      }
+    };
+  }
+});
+
+// src/api/outbound-sanitize.ts
+function sanitizeOutboundDiagnostics(value, parentKey = "") {
+  if (typeof value === "string") {
+    if (SANITIZED_VALUE_KEYS.test(parentKey) && /kernel/i.test(value)) {
+      return sanitizeVendorName(value);
+    }
+    return value;
+  }
+  if (Array.isArray(value)) return value.map((v) => sanitizeOutboundDiagnostics(v, parentKey));
+  if (value !== null && typeof value === "object") {
+    const out = {};
+    for (const [key, val] of Object.entries(value)) {
+      const renamed = KEY_RENAMES[key] ?? key;
+      out[renamed] = sanitizeOutboundDiagnostics(val, key);
+    }
+    return out;
+  }
+  return value;
+}
+function sanitizeAttempts(attempts) {
+  return attempts.map((a) => sanitizeOutboundDiagnostics(a));
+}
+function sanitizeHarvestResult(result) {
+  const diagnostics = result?.diagnostics;
+  if (!diagnostics?.debug) return result;
+  return {
+    ...result,
+    diagnostics: {
+      ...diagnostics,
+      debug: sanitizeOutboundDiagnostics(diagnostics.debug)
+    }
+  };
+}
+var KEY_RENAMES, SANITIZED_VALUE_KEYS;
+var init_outbound_sanitize = __esm({
+  "src/api/outbound-sanitize.ts"() {
+    "use strict";
+    init_errors();
+    KEY_RENAMES = {
+      kernel: "browserRuntime",
+      kernel_session_id: "browser_session_id",
+      kernel_delete_started: "session_cleanup_started",
+      kernel_delete_succeeded: "session_cleanup_succeeded",
+      kernel_delete_error: "session_cleanup_error",
+      kernelSessionId: "browserSessionId",
+      kernelDeleteStarted: "sessionCleanupStarted",
+      kernelDeleteSucceeded: "sessionCleanupSucceeded",
+      kernelDeleteError: "sessionCleanupError",
+      kernelProxyId: "proxyId"
+    };
+    SANITIZED_VALUE_KEYS = /error|message/i;
+  }
+});
+
 // src/blog/registry.ts
 var posts;
 var init_registry = __esm({
@@ -3425,7 +3528,7 @@ var init_url_utils = __esm({
 
 // src/api/kernel-fetch.ts
 async function fetchWithKernel(url) {
-  const apiKey = process.env.KERNEL_API_KEY;
+  const apiKey = browserServiceApiKey();
   if (!apiKey) throw new Error("Browser backend API key not set");
   const client = new import_sdk.default({ apiKey });
   const kb = await client.browsers.create({ stealth: true, timeout_seconds: 60 });
@@ -3450,6 +3553,7 @@ var init_kernel_fetch = __esm({
   "src/api/kernel-fetch.ts"() {
     "use strict";
     import_sdk = __toESM(require("@onkernel/sdk"), 1);
+    init_browser_service_env();
     import_playwright = require("playwright");
   }
 });
@@ -4239,8 +4343,8 @@ async function downloadAsset(url, destDir, filename) {
   }
   const writer = (0, import_node_fs.createWriteStream)(dest);
   await (0, import_promises2.pipeline)(import_node_stream.Readable.fromWeb(res.body), writer);
-  const { statSync } = await import("fs");
-  const sizeBytes = statSync(dest).size;
+  const { statSync: statSync2 } = await import("fs");
+  const sizeBytes = statSync2(dest).size;
   return { savedPath: dest, sizeBytes, mimeType };
 }
 async function harvestPageMedia(html, pageUrl, options = {}) {
@@ -8494,40 +8598,6 @@ var init_selectors = __esm({
   }
 });
 
-// src/errors.ts
-function sanitizeVendorName(message) {
-  return message.replace(/kernel\.sh\s+sessions?/gi, "sessions").replace(/kernel\.sh\s+session/gi, "this session").replace(/kernel\.sh/gi, "the service").replace(/kernel\s+sessions?/gi, "sessions").replace(/kernel\s+session/gi, "this session").replace(/\bkernel\b/gi, "the service").replace(/  +/g, " ").trim();
-}
-var RECAPTCHA_INSTRUCTIONS, CaptchaError, ExtractionError, RequestAbortedError;
-var init_errors = __esm({
-  "src/errors.ts"() {
-    "use strict";
-    RECAPTCHA_INSTRUCTIONS = "Google returned a CAPTCHA. Run with --headless=false to re-warm the browser profile, then retry.";
-    CaptchaError = class extends Error {
-      constructor(instructions) {
-        super(`CAPTCHA detected. ${instructions}`);
-        this.instructions = instructions;
-      }
-      instructions;
-      name = "CaptchaError";
-    };
-    ExtractionError = class extends Error {
-      constructor(message, cause) {
-        super(message);
-        this.cause = cause;
-      }
-      cause;
-      name = "ExtractionError";
-    };
-    RequestAbortedError = class extends Error {
-      name = "RequestAbortedError";
-      constructor(message = "Request aborted before harvest completed") {
-        super(message);
-      }
-    };
-  }
-});
-
 // src/driver/BrowserDriver.ts
 function positiveIntFromEnv(name, fallback) {
   const raw = process.env[name];
@@ -9509,7 +9579,7 @@ async function writeOutputs(result, outputDir) {
   }
 }
 async function ytHarvest(rawOptions) {
-  const kernelApiKey = process.env.KERNEL_API_KEY;
+  const kernelApiKey = browserServiceApiKey();
   if (!kernelApiKey) {
     throw new Error("A browser backend API key is required \u2014 YouTube harvesting requires a stealth session.");
   }
@@ -9543,6 +9613,7 @@ var init_youtube_harvest = __esm({
   "src/youtube/youtube-harvest.ts"() {
     "use strict";
     import_node_fs2 = require("fs");
+    init_browser_service_env();
     import_node_path4 = __toESM(require("path"), 1);
     import_papaparse = __toESM(require("papaparse"), 1);
     init_schemas2();
@@ -9620,7 +9691,7 @@ function parseTimedtextXml(xml) {
   return results;
 }
 async function fetchViaKernelInnertube(videoId) {
-  const kernelApiKey = process.env.KERNEL_API_KEY;
+  const kernelApiKey = browserServiceApiKey();
   if (!kernelApiKey) return null;
   const driver = new BrowserDriver();
   const start = Date.now();
@@ -9763,7 +9834,7 @@ async function attemptKernelWhisper(videoId, kernelApiKey, falKey, start) {
   }
 }
 async function fetchViaKernelWhisper(videoId) {
-  const kernelApiKey = process.env.KERNEL_API_KEY;
+  const kernelApiKey = browserServiceApiKey();
   const falKey = process.env.FAL_KEY;
   if (!kernelApiKey || !falKey) return null;
   const start = Date.now();
@@ -9803,6 +9874,7 @@ var init_CaptionFetcher = __esm({
   "src/youtube/CaptionFetcher.ts"() {
     "use strict";
     init_BrowserDriver();
+    init_browser_service_env();
     import_client2 = require("@fal-ai/client");
     WHISPER_RECORD_SECONDS = 90;
   }
@@ -10044,6 +10116,7 @@ var init_screenshot_routes = __esm({
   "src/api/screenshot-routes.ts"() {
     "use strict";
     import_hono3 = require("hono");
+    init_browser_service_env();
     import_zod14 = require("zod");
     init_screenshot();
     init_api_auth();
@@ -10078,7 +10151,7 @@ var init_screenshot_routes = __esm({
         }
         const device2 = body.device === "mobile" ? "mobile" : "desktop";
         try {
-          const buf = await captureScreenshot(parsedFallback.href, process.env.KERNEL_API_KEY?.trim(), device2);
+          const buf = await captureScreenshot(parsedFallback.href, browserServiceApiKey(), device2);
           return new Response(new Uint8Array(buf), {
             status: 200,
             headers: {
@@ -10094,7 +10167,7 @@ var init_screenshot_routes = __esm({
       }
       const device = body.device === "mobile" ? "mobile" : "desktop";
       try {
-        const buf = await captureScreenshot(urlCheck.parsed.href, process.env.KERNEL_API_KEY?.trim(), device);
+        const buf = await captureScreenshot(urlCheck.parsed.href, browserServiceApiKey(), device);
         return new Response(new Uint8Array(buf), {
           status: 200,
           headers: {
@@ -11379,29 +11452,30 @@ function buildPageIntelUrl(body, country) {
   return `https://www.facebook.com/ads/library/?active_status=all&ad_type=all&country=${country}&q=${encodeURIComponent(body.query.trim())}&search_type=keyword_unordered`;
 }
 function kernelLaunchOpts() {
-  return { headless: true, kernelApiKey: process.env.KERNEL_API_KEY?.trim(), kernelProxyId: process.env.KERNEL_PROXY_ID?.trim(), viewport: { width: 1280, height: 900 }, locale: "en-US" };
+  return { headless: true, kernelApiKey: browserServiceApiKey(), kernelProxyId: browserServiceProxyId(), viewport: { width: 1280, height: 900 }, locale: "en-US" };
 }
 async function kernelLaunchOptsResidential() {
-  let proxyId = process.env.KERNEL_PROXY_ID?.trim();
+  let proxyId = browserServiceProxyId();
   try {
     const resolution2 = await resolveKernelProxyId({
-      kernelApiKey: process.env.KERNEL_API_KEY?.trim(),
+      kernelApiKey: browserServiceApiKey(),
       proxyMode: "location",
-      configuredKernelProxyId: process.env.KERNEL_PROXY_ID?.trim(),
+      configuredKernelProxyId: browserServiceProxyId(),
       location: "New York, NY",
       gl: "us"
     });
     if (resolution2.kernelProxyId) proxyId = resolution2.kernelProxyId;
   } catch {
-    proxyId = process.env.KERNEL_PROXY_ID?.trim();
+    proxyId = browserServiceProxyId();
   }
-  return { headless: true, kernelApiKey: process.env.KERNEL_API_KEY?.trim(), kernelProxyId: proxyId, viewport: { width: 1280, height: 900 }, locale: "en-US" };
+  return { headless: true, kernelApiKey: browserServiceApiKey(), kernelProxyId: proxyId, viewport: { width: 1280, height: 900 }, locale: "en-US" };
 }
 var import_hono4, import_zod15, import_client3, FacebookAdBodySchema, FacebookPageIntelBodySchema, FacebookTranscribeBodySchema, FacebookSearchBodySchema, FacebookMediaBodySchema, facebookAdApp, ALLOWED_MEDIA_HOSTS;
 var init_facebook_ad_routes = __esm({
   "src/api/facebook-ad-routes.ts"() {
     "use strict";
     import_hono4 = require("hono");
+    init_browser_service_env();
     import_zod15 = require("zod");
     init_db();
     init_rates();
@@ -14321,8 +14395,8 @@ async function harvest(rawOptions) {
   const onAttemptEvent = getAttemptLogSink(rawOptions);
   const requestedProxyMode = raw.proxyMode;
   const proxyMode = requestedProxyMode === "none" ? "none" : requestedProxyMode === "configured" ? "configured" : "location";
-  const kernelApiKey = typeof raw.kernelApiKey === "string" ? raw.kernelApiKey.trim() : process.env.KERNEL_API_KEY?.trim();
-  const configuredKernelProxyId = typeof raw.kernelProxyId === "string" ? raw.kernelProxyId.trim() : process.env.KERNEL_PROXY_ID?.trim();
+  const kernelApiKey = typeof raw.kernelApiKey === "string" ? raw.kernelApiKey.trim() : browserServiceApiKey();
+  const configuredKernelProxyId = typeof raw.kernelProxyId === "string" ? raw.kernelProxyId.trim() : browserServiceProxyId();
   const proxyOpts = {
     kernelApiKey,
     proxyMode,
@@ -14509,6 +14583,7 @@ var init_harvest = __esm({
   "src/harvest.ts"() {
     "use strict";
     init_schemas3();
+    init_browser_service_env();
     init_BrowserDriver();
     init_PAAExtractor();
     init_OutputSerializer();
@@ -14933,8 +15008,8 @@ async function captureSerpIntelligenceSnapshot(rawInput, runtimeOptions = {}) {
       debug,
       serpOnly: true,
       headless: runtimeOptions.headless ?? true,
-      kernelApiKey: runtimeOptions.kernelApiKey ?? process.env.KERNEL_API_KEY?.trim(),
-      kernelProxyId: runtimeOptions.kernelProxyId ?? process.env.KERNEL_PROXY_ID?.trim(),
+      kernelApiKey: runtimeOptions.kernelApiKey ?? browserServiceApiKey(),
+      kernelProxyId: runtimeOptions.kernelProxyId ?? browserServiceProxyId(),
       format: "json",
       outputDir: runtimeOptions.outputDir ?? "/tmp/serp-intelligence-output",
       signal: runtimeOptions.signal,
@@ -14945,7 +15020,7 @@ async function captureSerpIntelligenceSnapshot(rawInput, runtimeOptions = {}) {
     const pageSnapshotLimit = normalizePageSnapshotLimit(parsedInput);
     const pageSnapshotTargets = collectPageSnapshotTargets(harvestResult, pageSnapshotLimit);
     const pageSnapshotArtifacts = pageSnapshotTargets.length > 0 ? (await capturePageSnapshotsFn(pageSnapshotTargets, {
-      kernelApiKey: runtimeOptions.kernelApiKey ?? process.env.KERNEL_API_KEY?.trim(),
+      kernelApiKey: runtimeOptions.kernelApiKey ?? browserServiceApiKey(),
       timeoutMs: runtimeOptions.pageSnapshotTimeoutMs,
       maxConcurrency: runtimeOptions.pageSnapshotMaxConcurrency,
       debug,
@@ -14967,6 +15042,7 @@ var init_serp_capture_service = __esm({
   "src/serp-intelligence/serp-capture-service.ts"() {
     "use strict";
     init_harvest();
+    init_browser_service_env();
     init_harvest_problems();
     init_page_snapshot_extractor();
     init_schemas4();
@@ -15071,6 +15147,7 @@ var init_serp_intelligence_routes = __esm({
   "src/api/serp-intelligence-routes.ts"() {
     "use strict";
     import_hono6 = require("hono");
+    init_browser_service_env();
     init_page_snapshot_extractor();
     init_serp_capture_service();
     init_schemas4();
@@ -15103,8 +15180,8 @@ var init_serp_intelligence_routes = __esm({
       if (!ok) return c.json(insufficientBalanceResponse(balance_mc, cost), 402);
       try {
         const result = await captureSerpIntelligenceSnapshot(parsed.data, {
-          kernelApiKey: process.env.KERNEL_API_KEY?.trim(),
-          kernelProxyId: process.env.KERNEL_PROXY_ID?.trim(),
+          kernelApiKey: browserServiceApiKey(),
+          kernelProxyId: browserServiceProxyId(),
           signal: c.req.raw.signal,
           billing: { creditsUsed: cost / 1e3 }
         });
@@ -15159,7 +15236,7 @@ var init_serp_intelligence_routes = __esm({
       if (!ok) return c.json(insufficientBalanceResponse(balance_mc, cost), 402);
       try {
         const result = await capturePageSnapshots(targets, {
-          kernelApiKey: process.env.KERNEL_API_KEY?.trim(),
+          kernelApiKey: browserServiceApiKey(),
           timeoutMs: parsed.data.timeoutMs,
           maxConcurrency: parsed.data.maxConcurrency,
           debug: parsed.data.debug
@@ -15199,207 +15276,7 @@ var PACKAGE_VERSION;
 var init_version = __esm({
   "src/version.ts"() {
     "use strict";
-    PACKAGE_VERSION = "0.1.7";
-  }
-});
-
-// src/mcp/mcp-tool-schemas.ts
-var import_zod19, HarvestPaaInputSchema, ExtractUrlInputSchema, MapSiteUrlsInputSchema, ExtractSiteInputSchema, YoutubeHarvestInputSchema, YoutubeTranscribeInputSchema, FacebookPageIntelInputSchema, FacebookAdSearchInputSchema, FacebookAdTranscribeInputSchema, MapsPlaceIntelInputSchema, MapsSearchInputSchema, NullableString, MapsSearchOutputSchema, MapSiteUrlsOutputSchema, YoutubeHarvestOutputSchema, FacebookAdSearchOutputSchema, FacebookPageIntelOutputSchema, CreditsInfoInputSchema, SearchSerpInputSchema, CaptureSerpSnapshotInputSchema, ScreenshotInputSchema, CaptureSerpPageSnapshotsInputSchema;
-var init_mcp_tool_schemas = __esm({
-  "src/mcp/mcp-tool-schemas.ts"() {
-    "use strict";
-    import_zod19 = require("zod");
-    HarvestPaaInputSchema = {
-      query: import_zod19.z.string().min(1).describe('Core search topic only. If the user says "best hvac company in Denver CO", use query="best hvac company" and location="Denver, CO". Do not include the location in query when it can be separated.'),
-      location: import_zod19.z.string().optional().describe('City, region, or country for geo-targeted results, inferred from the user request when present, e.g. "Denver, CO", "Tokyo, Japan", "London, UK".'),
-      maxQuestions: import_zod19.z.number().int().min(1).max(200).default(30).describe("Number of PAA questions to extract. Default 30. Maximum 200. Use 10 for quick probes, 30 for normal research, 100-200 when the user asks for everything/full/deep research. Larger harvests get a longer server time budget (151-200 questions \u2192 up to 280s). Credits are charged by extracted question; unused request hold is refunded."),
-      gl: import_zod19.z.string().length(2).default("us").describe("Google country code inferred from location or user language. Examples: United States us, United Kingdom gb, Japan jp, Canada ca, Australia au."),
-      hl: import_zod19.z.string().default("en").describe("Google interface/content language inferred from the user request. Use en unless the user asks for another language or locale."),
-      device: import_zod19.z.enum(["desktop", "mobile"]).default("desktop").describe("SERP device context. Use desktop by default; use mobile only when the user asks for mobile rankings."),
-      proxyMode: import_zod19.z.enum(["location", "configured", "none"]).default("location").describe("Proxy targeting mode. Use location by default so city/state searches create or reuse a matching residential proxy. Use configured for the static configured proxy. Use none only for direct-network debugging."),
-      proxyZip: import_zod19.z.string().regex(/^\d{5}$/).optional().describe("Optional US ZIP override for residential location proxy targeting. Use only when the user gives a specific ZIP or city-center proxy targeting needs to be forced."),
-      debug: import_zod19.z.boolean().default(false).describe("Include sanitized browser/session/location diagnostics in the response. Use true when debugging localization, CAPTCHA, or proxy behavior.")
-    };
-    ExtractUrlInputSchema = {
-      url: import_zod19.z.string().url().describe("Public http/https URL to extract. Use this when the user provides one specific page URL."),
-      screenshot: import_zod19.z.boolean().default(false).describe("Also capture a full-page screenshot of the URL. Saved to ~/Downloads/mcp-scraper/screenshots/ and returned inline. Use when the user asks to see or capture the page visually."),
-      screenshotDevice: import_zod19.z.enum(["desktop", "mobile"]).default("desktop").describe("Viewport for screenshot. desktop = 1440\xD7900. mobile = 390\xD7844. Default desktop."),
-      extractBranding: import_zod19.z.boolean().default(false).describe("Extract brand colors, fonts, logo, and favicon using a rendered browser session. Returns colorScheme (light/dark), colors (primary/accent/background/text/heading as hex), fonts (heading/body family names), and assets (logo URL, favicon URL). Use when the user asks about brand colors, site theme, or brand assets."),
-      downloadMedia: import_zod19.z.boolean().default(false).describe("Extract and download all page media (images, video, audio) to ~/Downloads/mcp-scraper/media/. Ad networks, tracking pixels, and noise URLs are filtered automatically. Use when the user asks to download or harvest assets from a page."),
-      mediaTypes: import_zod19.z.array(import_zod19.z.enum(["image", "video", "audio"])).default(["image", "video", "audio"]).describe("Which media types to download. Default all three."),
-      allowLocal: import_zod19.z.boolean().default(false).describe("Allow localhost and private-network URLs. For local development only.")
-    };
-    MapSiteUrlsInputSchema = {
-      url: import_zod19.z.string().url().describe("Public website URL or domain to crawl for internal URLs. Use before extract_site when the user asks to audit/map/crawl a site."),
-      maxUrls: import_zod19.z.number().int().min(1).max(500).optional().describe("Maximum URLs to discover. Use 100 for normal maps, higher when the user asks for a full inventory.")
-    };
-    ExtractSiteInputSchema = {
-      url: import_zod19.z.string().url().describe("Public website URL or domain to extract across multiple pages. Use when the user asks for a site audit, website crawl, or full-site content/schema extraction."),
-      maxPages: import_zod19.z.number().int().min(1).max(50).optional().describe("Maximum pages to extract. Use 50 when the user asks for full results or a complete crawl within MCP limits.")
-    };
-    YoutubeHarvestInputSchema = {
-      mode: import_zod19.z.enum(["search", "channel"]).describe("Use search for topic/keyword requests. Use channel when the user provides @handle, channel ID, or channel URL."),
-      query: import_zod19.z.string().optional().describe("Required when mode is search. The YouTube search topic in the user\u2019s words."),
-      channelHandle: import_zod19.z.string().optional().describe("YouTube channel handle, channel ID, or URL. Examples: @mkbhd, UC..., https://youtube.com/@mkbhd."),
-      maxVideos: import_zod19.z.number().int().min(1).max(500).default(50).describe("Number of videos to return. Default 50. Increase when user asks for full channel/history.")
-    };
-    YoutubeTranscribeInputSchema = {
-      videoId: import_zod19.z.string().min(1).describe("YouTube video ID, e.g. dQw4w9WgXcQ")
-    };
-    FacebookPageIntelInputSchema = {
-      pageId: import_zod19.z.string().optional(),
-      libraryId: import_zod19.z.string().optional(),
-      query: import_zod19.z.string().optional().describe("Advertiser or brand name when pageId/libraryId is not known. One of pageId, libraryId, or query is required."),
-      maxAds: import_zod19.z.number().int().min(1).max(200).default(50),
-      country: import_zod19.z.string().length(2).default("US")
-    };
-    FacebookAdSearchInputSchema = {
-      query: import_zod19.z.string().min(1).describe("Advertiser, brand, competitor, niche, or keyword to search in Facebook Ad Library."),
-      country: import_zod19.z.string().length(2).default("US"),
-      maxResults: import_zod19.z.number().int().min(1).max(20).default(10)
-    };
-    FacebookAdTranscribeInputSchema = {
-      videoUrl: import_zod19.z.string().url().describe("Facebook CDN video URL from a facebook_page_intel result")
-    };
-    MapsPlaceIntelInputSchema = {
-      businessName: import_zod19.z.string().min(1).describe('Business name only. If user says "Elite Roofing Denver CO", use businessName="Elite Roofing" and location="Denver, CO".'),
-      location: import_zod19.z.string().min(1).describe('City/region/country where the business should be searched, e.g. "Denver, CO". Infer from the user request when possible.'),
-      gl: import_zod19.z.string().length(2).default("us").describe("Google country code inferred from location."),
-      hl: import_zod19.z.string().length(2).default("en").describe("Language inferred from user request."),
-      includeReviews: import_zod19.z.boolean().default(false).describe("Whether to fetch individual review cards"),
-      maxReviews: import_zod19.z.number().int().min(1).max(500).default(50).describe("Max review cards to return (requires includeReviews: true)")
-    };
-    MapsSearchInputSchema = {
-      query: import_zod19.z.string().min(1).describe('Business category, niche, keyword, or search term. If the user says "roofers in Denver CO", use query="roofers" and location="Denver, CO". Do not put the location here when it can be separated.'),
-      location: import_zod19.z.string().optional().describe('City, region, country, or service area for the Maps search, e.g. "Denver, CO". Infer from the user request when present.'),
-      gl: import_zod19.z.string().length(2).default("us").describe("Google country code inferred from location."),
-      hl: import_zod19.z.string().length(2).default("en").describe("Language inferred from user request."),
-      maxResults: import_zod19.z.number().int().min(1).max(50).default(10).describe("Number of Google Maps business/profile candidates to return. Default 10. Maximum 50. Use 10 unless the user asks for more.")
-    };
-    NullableString = import_zod19.z.string().nullable();
-    MapsSearchOutputSchema = {
-      query: import_zod19.z.string(),
-      location: import_zod19.z.string().nullable(),
-      searchQuery: import_zod19.z.string(),
-      searchUrl: import_zod19.z.string().url(),
-      extractedAt: import_zod19.z.string(),
-      requestedMaxResults: import_zod19.z.number().int().min(1).max(50),
-      resultCount: import_zod19.z.number().int().min(0).max(50),
-      results: import_zod19.z.array(import_zod19.z.object({
-        position: import_zod19.z.number().int().min(1),
-        name: import_zod19.z.string(),
-        placeUrl: import_zod19.z.string().url(),
-        cid: NullableString,
-        cidDecimal: NullableString,
-        rating: NullableString,
-        reviewCount: NullableString,
-        category: NullableString,
-        address: NullableString,
-        websiteUrl: NullableString,
-        directionsUrl: NullableString,
-        metadata: import_zod19.z.array(import_zod19.z.string())
-      })),
-      durationMs: import_zod19.z.number().int().min(0)
-    };
-    MapSiteUrlsOutputSchema = {
-      startUrl: import_zod19.z.string(),
-      totalFound: import_zod19.z.number().int().min(0),
-      truncated: import_zod19.z.boolean(),
-      okCount: import_zod19.z.number().int().min(0),
-      redirectCount: import_zod19.z.number().int().min(0),
-      brokenCount: import_zod19.z.number().int().min(0),
-      urls: import_zod19.z.array(import_zod19.z.object({
-        url: import_zod19.z.string(),
-        status: import_zod19.z.number().int().nullable()
-      })),
-      durationMs: import_zod19.z.number().min(0)
-    };
-    YoutubeHarvestOutputSchema = {
-      mode: import_zod19.z.string(),
-      videoCount: import_zod19.z.number().int().min(0),
-      channel: import_zod19.z.object({
-        title: NullableString,
-        subscriberCount: NullableString
-      }).nullable(),
-      videos: import_zod19.z.array(import_zod19.z.object({
-        videoId: import_zod19.z.string(),
-        title: import_zod19.z.string(),
-        channelName: NullableString,
-        views: NullableString,
-        duration: NullableString,
-        url: NullableString
-      }))
-    };
-    FacebookAdSearchOutputSchema = {
-      query: import_zod19.z.string(),
-      advertiserCount: import_zod19.z.number().int().min(0),
-      advertisers: import_zod19.z.array(import_zod19.z.object({
-        name: NullableString,
-        adCount: import_zod19.z.number().int().nullable(),
-        libraryId: NullableString
-      }))
-    };
-    FacebookPageIntelOutputSchema = {
-      advertiserName: NullableString,
-      totalAds: import_zod19.z.number().int().min(0),
-      activeCount: import_zod19.z.number().int().min(0),
-      videoCount: import_zod19.z.number().int().min(0),
-      imageCount: import_zod19.z.number().int().min(0),
-      ads: import_zod19.z.array(import_zod19.z.object({
-        libraryId: NullableString,
-        status: NullableString,
-        creativeType: NullableString,
-        headline: NullableString,
-        cta: NullableString,
-        startDate: NullableString,
-        videoUrl: NullableString,
-        variations: import_zod19.z.number().int().nullable()
-      }))
-    };
-    CreditsInfoInputSchema = {
-      item: import_zod19.z.string().optional().describe('Optional tool, action, or feature to look up, e.g. "maps reviews", "extract_url", or "YouTube transcription"'),
-      includeLedger: import_zod19.z.boolean().default(false).describe("Whether to include recent credit ledger entries")
-    };
-    SearchSerpInputSchema = {
-      query: import_zod19.z.string().min(1).describe('Core search topic only. Separate location when possible. If user says "best dentist in Brooklyn NY serp", use query="best dentist" and location="Brooklyn, NY".'),
-      location: import_zod19.z.string().optional().describe("City, region, or country for geo-targeted results, inferred from user request when present."),
-      gl: import_zod19.z.string().length(2).default("us").describe("Google country code inferred from location or user language."),
-      hl: import_zod19.z.string().default("en").describe("Google interface/content language inferred from user request."),
-      device: import_zod19.z.enum(["desktop", "mobile"]).default("desktop").describe("SERP device context. Use desktop by default; use mobile only when the user asks for mobile rankings."),
-      proxyMode: import_zod19.z.enum(["location", "configured", "none"]).default("location").describe("Proxy targeting mode. Use location by default so city/state searches create or reuse a matching residential proxy. Use configured for the static configured proxy. Use none only for direct-network debugging."),
-      proxyZip: import_zod19.z.string().regex(/^\d{5}$/).optional().describe("Optional US ZIP override for residential location proxy targeting. Use only when the user gives a specific ZIP or city-center proxy targeting needs to be forced."),
-      debug: import_zod19.z.boolean().default(false).describe("Include sanitized browser/session/location diagnostics in the response. Use true when debugging localization, CAPTCHA, or proxy behavior."),
-      pages: import_zod19.z.number().int().min(1).max(2).default(1).describe("Number of result pages to fetch (1\u20132)")
-    };
-    CaptureSerpSnapshotInputSchema = {
-      query: import_zod19.z.string().min(1).describe("Core search query to capture as a structured SERP Intelligence snapshot. Separate the place into location when the user gives a city, region, country, or ZIP."),
-      location: import_zod19.z.string().optional().describe("City, region, country, or service area used for localized Google results. MCP Scraper records location evidence; UULE alone is not proof of localization."),
-      gl: import_zod19.z.string().length(2).default("us").describe("Google country code inferred from the requested market, e.g. us, gb, ca, au."),
-      hl: import_zod19.z.string().default("en").describe("Google interface/content language inferred from the user request."),
-      device: import_zod19.z.enum(["desktop", "mobile"]).default("desktop").describe("SERP device context. Use mobile only when the user asks for mobile rankings or mobile SERP evidence."),
-      proxyMode: import_zod19.z.enum(["location", "configured", "none"]).default("location").describe("Proxy behavior for capture. Use location for localized residential proxy targeting, configured for the static residential proxy, and none only for direct-network debugging."),
-      proxyZip: import_zod19.z.string().regex(/^\d{5}$/).optional().describe("Optional US ZIP override for residential location proxy targeting when a precise city-center or ZIP proxy is needed."),
-      pages: import_zod19.z.number().int().min(1).max(2).default(1).describe("Number of Google result pages to capture. Use 1 normally and 2 only when the user needs deeper ranking evidence."),
-      debug: import_zod19.z.boolean().default(false).describe("Include sanitized browser, proxy, and location diagnostics. Use true when debugging localization, CAPTCHA, proxy selection, or capture reliability."),
-      includePageSnapshots: import_zod19.z.boolean().default(false).describe("Also capture ranking-page snapshots for selected SERP URLs through the same product capture path."),
-      pageSnapshotLimit: import_zod19.z.number().int().min(0).max(10).default(0).describe("Maximum ranking-page snapshots to capture when includePageSnapshots is true. Use 0 when only SERP evidence is needed.")
-    };
-    ScreenshotInputSchema = {
-      url: import_zod19.z.string().url().describe("URL to capture as a full-page screenshot. Use http or https. Pass allowLocal: true to capture localhost or private-network URLs during development."),
-      device: import_zod19.z.enum(["desktop", "mobile"]).default("desktop").describe("Viewport profile. desktop = 1440\xD7900. mobile = 390\xD7844. Use desktop by default; use mobile when the user asks for a mobile view."),
-      allowLocal: import_zod19.z.boolean().default(false).describe("Allow localhost and private-network URLs (127.x, 192.168.x, 10.x, etc.). For local development only \u2014 not for production use.")
-    };
-    CaptureSerpPageSnapshotsInputSchema = {
-      urls: import_zod19.z.array(import_zod19.z.string().url()).min(1).max(25).describe("Public HTTP/HTTPS URLs to capture as SERP Intelligence page snapshots. Do not pass localhost, private IPs, file URLs, or internal admin URLs."),
-      targets: import_zod19.z.array(import_zod19.z.object({
-        url: import_zod19.z.string().url().describe("Public HTTP/HTTPS URL to capture."),
-        sourceKind: import_zod19.z.enum(["organic", "ai_citation", "local_pack_website", "configured_target", "site_subject"]).default("configured_target").describe("Why this page is being captured for SERP Intelligence evidence."),
-        sourcePosition: import_zod19.z.number().int().min(1).optional().describe("Ranking or citation position when the page came from SERP evidence.")
-      }).strict()).min(1).max(25).optional().describe("Structured page snapshot targets. Use this instead of urls when source kind or position should be preserved."),
-      maxConcurrency: import_zod19.z.number().int().min(1).max(5).default(2).describe("Parallel page captures. Use 2 normally; higher values can increase proxy/browser pressure."),
-      timeoutMs: import_zod19.z.number().int().min(1e3).max(6e4).default(15e3).describe("Per-page capture timeout in milliseconds. Increase for slow pages; timeout artifacts are returned as structured capture failures."),
-      debug: import_zod19.z.boolean().default(false).describe("Include sanitized browser/proxy diagnostics for page snapshot debugging. Use true for capture, network, or proxy troubleshooting.")
-    };
+    PACKAGE_VERSION = "0.1.9";
   }
 });
 
@@ -15505,7 +15382,7 @@ function debugSection(debug) {
   if (!debug || typeof debug !== "object") return "";
   const request = debug.request ?? {};
   const browser = debug.browser ?? {};
-  const kernel = browser.kernel ?? {};
+  const kernel = browser.browserRuntime ?? browser.kernel ?? {};
   const network = browser.networkLocation ?? {};
   const nav = browser.serpNavigation ?? {};
   const proxyResolution = kernel.proxyResolution ?? {};
@@ -15531,12 +15408,14 @@ function errorAttemptsSection(body) {
   const lines = attempts.slice(0, 5).map((attempt) => {
     const debug = attempt.debug ?? {};
     const browser = debug.browser ?? {};
-    const kernel = browser.kernel ?? {};
+    const kernel = browser.browserRuntime ?? browser.kernel ?? {};
     const proxyResolution = kernel.proxyResolution ?? {};
     const network = browser.networkLocation ?? {};
     const nav = browser.serpNavigation ?? {};
     const geo = [network.ip, network.city, network.region].filter(Boolean).join(" / ") || "geo unknown";
-    return `- Attempt ${attempt.attempt_number ?? "?"}: ${attempt.outcome ?? attempt.status ?? "unknown"} \xB7 session ${attempt.kernel_session_id ?? kernel.sessionId ?? "unknown"} \xB7 proxy ${debug.request?.proxyMode ?? kernel.proxyMode ?? "unknown"}${proxyResolution.source ? `/${proxyResolution.source}` : ""} \xB7 ${geo} \xB7 CAPTCHA ${nav.captchaDetected === true ? "yes" : nav.captchaDetected === false ? "no" : "unknown"} \xB7 deleted ${attempt.kernel_delete_succeeded === true ? "yes" : attempt.kernel_delete_succeeded === false ? "no" : "unknown"}`;
+    const sessionId = attempt.browser_session_id ?? attempt.kernel_session_id ?? kernel.sessionId ?? "unknown";
+    const cleanupSucceeded2 = attempt.session_cleanup_succeeded ?? attempt.kernel_delete_succeeded;
+    return `- Attempt ${attempt.attempt_number ?? "?"}: ${attempt.outcome ?? attempt.status ?? "unknown"} \xB7 session ${sessionId} \xB7 proxy ${debug.request?.proxyMode ?? kernel.proxyMode ?? "unknown"}${proxyResolution.source ? `/${proxyResolution.source}` : ""} \xB7 ${geo} \xB7 CAPTCHA ${nav.captchaDetected === true ? "yes" : nav.captchaDetected === false ? "no" : "unknown"} \xB7 cleanup ${cleanupSucceeded2 === true ? "yes" : cleanupSucceeded2 === false ? "no" : "unknown"}`;
   });
   return `
 
@@ -15583,7 +15462,31 @@ ${serpRows}` : "";
   const full = `# PAA Report: "${input.query}"${input.location ? ` \xB7 ${input.location}` : ""}
 
 ${paaTable}${serpTable}${entityIdsSection(entityIds)}${aiSection}${statsLine}${debugSection(diagnostics?.debug)}${tips}`;
-  return oneBlock(full);
+  return {
+    ...oneBlock(full),
+    structuredContent: {
+      query: input.query,
+      location: input.location ?? null,
+      questionCount: flat.length,
+      completionStatus: diagnostics?.completionStatus ?? null,
+      questions: flat.map((r) => ({
+        question: String(r.question ?? ""),
+        answer: r.answer ?? null,
+        sourceTitle: r.source_title ?? null,
+        sourceSite: r.source_site ?? null
+      })),
+      organicResults: organic.map((r) => ({
+        position: Number(r.position) || 0,
+        title: String(r.title ?? ""),
+        url: String(r.url ?? ""),
+        domain: String(r.domain ?? ""),
+        snippet: r.snippet ?? null
+      })),
+      aiOverview: aiOvw ? { detected: aiOvw.detected === true, text: aiOvw.text ?? null } : null,
+      entityIds: entityIds ? { kgIds: entityIds.kgIds ?? [], cids: entityIds.cids ?? [], gcids: entityIds.gcids ?? [] } : null,
+      durationMs: durationMs ?? null
+    }
+  };
 }
 function formatSearchSerp(raw, input) {
   const parsed = parseData(raw);
@@ -15621,7 +15524,29 @@ ${localRows}` : "";
   const full = `# SERP Report: "${input.query}"${input.location ? ` \xB7 ${input.location}` : ""}
 
 ${serpTable}${localSection}${entityIdsSection(entityIds)}${aiSection}${debugSection(diagnostics?.debug)}${tips}`;
-  return oneBlock(full);
+  return {
+    ...oneBlock(full),
+    structuredContent: {
+      query: input.query,
+      location: input.location ?? null,
+      organicResults: organic.map((r) => ({
+        position: Number(r.position) || 0,
+        title: String(r.title ?? ""),
+        url: String(r.url ?? ""),
+        domain: String(r.domain ?? ""),
+        snippet: r.snippet ?? null
+      })),
+      localPack: localPack.map((b) => ({
+        position: Number(b.position) || 0,
+        name: String(b.name ?? ""),
+        rating: b.rating ?? null,
+        reviewCount: b.reviewCount ?? null,
+        websiteUrl: b.websiteUrl ?? null
+      })),
+      aiOverview: aiOvw ? { detected: aiOvw.detected === true, text: aiOvw.text ?? null } : null,
+      entityIds: entityIds ? { kgIds: entityIds.kgIds ?? [], cids: entityIds.cids ?? [], gcids: entityIds.gcids ?? [] } : null
+    }
+  };
 }
 function formatExtractUrl(raw, input) {
   const parsed = parseData(raw);
@@ -15690,15 +15615,27 @@ ${bodyMd.slice(0, 3e3)}${bodyMd.length > 3e3 ? "\n\n*(truncated)*" : ""}` : "";
 **${title}**
 ${headingSection}${kpoSection}${brandingSection}${bodySection}${screenshotSection}${mediaSection}${tips}`;
   const textResult = oneBlock(full);
+  const structuredContent = {
+    url,
+    title: d.title ?? null,
+    headings: headings.map((h) => ({ level: Number(h.level) || 0, text: String(h.text ?? "") })),
+    schemaBlockCount: schemaCount,
+    entityName: kpo?.entityName ?? null,
+    entityTypes: kpo?.type ?? [],
+    napScore: kpo?.napScore ?? null,
+    missingSchemaFields: kpo?.missingFields ?? [],
+    screenshotSaved: screenshotPath ?? null
+  };
   if (screenshotMeta?.base64) {
     return {
       content: [
         ...textResult.content,
         { type: "image", data: screenshotMeta.base64, mimeType: "image/png" }
-      ]
+      ],
+      structuredContent
     };
   }
-  return textResult;
+  return { ...textResult, structuredContent };
 }
 function formatMapSiteUrls(raw, input) {
   const parsed = parseData(raw);
@@ -15768,7 +15705,19 @@ ${pageRows}`,
 - Map URLs first: use \`map_site_urls\`
 - Inspect a single page: use \`extract_url\``
   ].join("\n");
-  return oneBlock(full);
+  return {
+    ...oneBlock(full),
+    structuredContent: {
+      url: input.url,
+      pageCount: pages.length,
+      pages: pages.map((p) => ({
+        url: String(p.url ?? ""),
+        title: p.title ?? null,
+        schemaTypes: p.kpo?.type ?? []
+      })),
+      durationMs: d.durationMs ?? 0
+    }
+  };
 }
 function formatYoutubeHarvest(raw, input) {
   const parsed = parseData(raw);
@@ -15965,7 +15914,26 @@ ${costRows}` : "",
 |------|-----------|---------|-------------|
 ${ledgerRows}` : ""
   ].filter(Boolean).join("\n");
-  return oneBlock(full);
+  return {
+    ...oneBlock(full),
+    structuredContent: {
+      balanceCredits: typeof balance === "number" ? balance : null,
+      matchedCost: matched ? { label: matched.label, credits: matched.credits, unit: matched.unit, notes: matched.notes ?? null } : null,
+      costs: costs.map((c) => ({
+        key: c.key,
+        label: c.label,
+        credits: c.credits,
+        unit: c.unit,
+        notes: c.notes ?? null
+      })),
+      ledger: ledger.map((row) => ({
+        createdAt: String(row.created_at ?? ""),
+        operation: String(row.operation ?? ""),
+        credits: row.amount_mc / 1e3,
+        description: row.description ?? null
+      }))
+    }
+  };
 }
 function formatMapsSearch(raw, input) {
   const parsed = parseData(raw);
@@ -16114,7 +16082,28 @@ ${entitySection}` : null,
 ---
 *Extracted in ${(durationMs / 1e3).toFixed(1)}s*` : null
   ].filter(Boolean).join("\n");
-  return oneBlock(full);
+  return {
+    ...oneBlock(full),
+    structuredContent: {
+      name,
+      rating: rating ?? null,
+      reviewCount: reviewCount ?? null,
+      category: category ?? null,
+      address: address ?? null,
+      phone: phone ?? null,
+      website: website ?? null,
+      hoursSummary: hoursSummary ?? null,
+      bookingUrl: bookingUrl ?? null,
+      kgmid: kgmid ?? null,
+      cidDecimal: cidDecimal ?? null,
+      cidUrl: cidUrl ?? null,
+      lat: lat ?? null,
+      lng: lng ?? null,
+      reviewsStatus,
+      reviewsCollected: reviews.length,
+      reviewTopics: topics.map((t) => ({ label: String(t.label ?? ""), count: String(t.count ?? "") }))
+    }
+  };
 }
 function formatFacebookAdTranscribe(raw, input) {
   const parsed = parseData(raw);
@@ -16158,6 +16147,320 @@ var init_mcp_response_formatter = __esm({
   }
 });
 
+// src/mcp/mcp-tool-schemas.ts
+var import_zod19, HarvestPaaInputSchema, ExtractUrlInputSchema, MapSiteUrlsInputSchema, ExtractSiteInputSchema, YoutubeHarvestInputSchema, YoutubeTranscribeInputSchema, FacebookPageIntelInputSchema, FacebookAdSearchInputSchema, FacebookAdTranscribeInputSchema, MapsPlaceIntelInputSchema, MapsSearchInputSchema, NullableString, MapsSearchOutputSchema, OrganicResultOutput, AiOverviewOutput, EntityIdsOutput, HarvestPaaOutputSchema, SearchSerpOutputSchema, ExtractUrlOutputSchema, ExtractSiteOutputSchema, MapsPlaceIntelOutputSchema, CreditsInfoOutputSchema, MapSiteUrlsOutputSchema, YoutubeHarvestOutputSchema, FacebookAdSearchOutputSchema, FacebookPageIntelOutputSchema, CreditsInfoInputSchema, SearchSerpInputSchema, CaptureSerpSnapshotInputSchema, ScreenshotInputSchema, CaptureSerpPageSnapshotsInputSchema;
+var init_mcp_tool_schemas = __esm({
+  "src/mcp/mcp-tool-schemas.ts"() {
+    "use strict";
+    import_zod19 = require("zod");
+    HarvestPaaInputSchema = {
+      query: import_zod19.z.string().min(1).describe('Core search topic only. If the user says "best hvac company in Denver CO", use query="best hvac company" and location="Denver, CO". Do not include the location in query when it can be separated.'),
+      location: import_zod19.z.string().optional().describe('City, region, or country for geo-targeted results, inferred from the user request when present, e.g. "Denver, CO", "Tokyo, Japan", "London, UK".'),
+      maxQuestions: import_zod19.z.number().int().min(1).max(200).default(30).describe("Number of PAA questions to extract. Default 30. Maximum 200. Use 10 for quick probes, 30 for normal research, 100-200 when the user asks for everything/full/deep research. Larger harvests get a longer server time budget (151-200 questions \u2192 up to 280s). Credits are charged by extracted question; unused request hold is refunded."),
+      gl: import_zod19.z.string().length(2).default("us").describe("Google country code inferred from location or user language. Examples: United States us, United Kingdom gb, Japan jp, Canada ca, Australia au."),
+      hl: import_zod19.z.string().default("en").describe("Google interface/content language inferred from the user request. Use en unless the user asks for another language or locale."),
+      device: import_zod19.z.enum(["desktop", "mobile"]).default("desktop").describe("SERP device context. Use desktop by default; use mobile only when the user asks for mobile rankings."),
+      proxyMode: import_zod19.z.enum(["location", "configured", "none"]).default("location").describe("Proxy targeting mode. Use location by default so city/state searches create or reuse a matching residential proxy. Use configured for the static configured proxy. Use none only for direct-network debugging."),
+      proxyZip: import_zod19.z.string().regex(/^\d{5}$/).optional().describe("Optional US ZIP override for residential location proxy targeting. Use only when the user gives a specific ZIP or city-center proxy targeting needs to be forced."),
+      debug: import_zod19.z.boolean().default(false).describe("Include sanitized browser/session/location diagnostics in the response. Use true when debugging localization, CAPTCHA, or proxy behavior.")
+    };
+    ExtractUrlInputSchema = {
+      url: import_zod19.z.string().url().describe("Public http/https URL to extract. Use this when the user provides one specific page URL."),
+      screenshot: import_zod19.z.boolean().default(false).describe("Also capture a full-page screenshot of the URL. Saved to ~/Downloads/mcp-scraper/screenshots/ and returned inline. Use when the user asks to see or capture the page visually."),
+      screenshotDevice: import_zod19.z.enum(["desktop", "mobile"]).default("desktop").describe("Viewport for screenshot. desktop = 1440\xD7900. mobile = 390\xD7844. Default desktop."),
+      extractBranding: import_zod19.z.boolean().default(false).describe("Extract brand colors, fonts, logo, and favicon using a rendered browser session. Returns colorScheme (light/dark), colors (primary/accent/background/text/heading as hex), fonts (heading/body family names), and assets (logo URL, favicon URL). Use when the user asks about brand colors, site theme, or brand assets."),
+      downloadMedia: import_zod19.z.boolean().default(false).describe("Extract and download all page media (images, video, audio) to ~/Downloads/mcp-scraper/media/. Ad networks, tracking pixels, and noise URLs are filtered automatically. Use when the user asks to download or harvest assets from a page."),
+      mediaTypes: import_zod19.z.array(import_zod19.z.enum(["image", "video", "audio"])).default(["image", "video", "audio"]).describe("Which media types to download. Default all three."),
+      allowLocal: import_zod19.z.boolean().default(false).describe("Allow localhost and private-network URLs. For local development only.")
+    };
+    MapSiteUrlsInputSchema = {
+      url: import_zod19.z.string().url().describe("Public website URL or domain to crawl for internal URLs. Use before extract_site when the user asks to audit/map/crawl a site."),
+      maxUrls: import_zod19.z.number().int().min(1).max(500).optional().describe("Maximum URLs to discover. Use 100 for normal maps, higher when the user asks for a full inventory.")
+    };
+    ExtractSiteInputSchema = {
+      url: import_zod19.z.string().url().describe("Public website URL or domain to extract across multiple pages. Use when the user asks for a site audit, website crawl, or full-site content/schema extraction."),
+      maxPages: import_zod19.z.number().int().min(1).max(50).optional().describe("Maximum pages to extract. Use 50 when the user asks for full results or a complete crawl within MCP limits.")
+    };
+    YoutubeHarvestInputSchema = {
+      mode: import_zod19.z.enum(["search", "channel"]).describe("Use search for topic/keyword requests. Use channel when the user provides @handle, channel ID, or channel URL."),
+      query: import_zod19.z.string().optional().describe("Required when mode is search. The YouTube search topic in the user\u2019s words."),
+      channelHandle: import_zod19.z.string().optional().describe("YouTube channel handle, channel ID, or URL. Examples: @mkbhd, UC..., https://youtube.com/@mkbhd."),
+      maxVideos: import_zod19.z.number().int().min(1).max(500).default(50).describe("Number of videos to return. Default 50. Increase when user asks for full channel/history.")
+    };
+    YoutubeTranscribeInputSchema = {
+      videoId: import_zod19.z.string().min(1).describe("YouTube video ID, e.g. dQw4w9WgXcQ")
+    };
+    FacebookPageIntelInputSchema = {
+      pageId: import_zod19.z.string().optional(),
+      libraryId: import_zod19.z.string().optional(),
+      query: import_zod19.z.string().optional().describe("Advertiser or brand name when pageId/libraryId is not known. One of pageId, libraryId, or query is required."),
+      maxAds: import_zod19.z.number().int().min(1).max(200).default(50),
+      country: import_zod19.z.string().length(2).default("US")
+    };
+    FacebookAdSearchInputSchema = {
+      query: import_zod19.z.string().min(1).describe("Advertiser, brand, competitor, niche, or keyword to search in Facebook Ad Library."),
+      country: import_zod19.z.string().length(2).default("US"),
+      maxResults: import_zod19.z.number().int().min(1).max(20).default(10)
+    };
+    FacebookAdTranscribeInputSchema = {
+      videoUrl: import_zod19.z.string().url().describe("Facebook CDN video URL from a facebook_page_intel result")
+    };
+    MapsPlaceIntelInputSchema = {
+      businessName: import_zod19.z.string().min(1).describe('Business name only. If user says "Elite Roofing Denver CO", use businessName="Elite Roofing" and location="Denver, CO".'),
+      location: import_zod19.z.string().min(1).describe('City/region/country where the business should be searched, e.g. "Denver, CO". Infer from the user request when possible.'),
+      gl: import_zod19.z.string().length(2).default("us").describe("Google country code inferred from location."),
+      hl: import_zod19.z.string().length(2).default("en").describe("Language inferred from user request."),
+      includeReviews: import_zod19.z.boolean().default(false).describe("Whether to fetch individual review cards"),
+      maxReviews: import_zod19.z.number().int().min(1).max(500).default(50).describe("Max review cards to return (requires includeReviews: true)")
+    };
+    MapsSearchInputSchema = {
+      query: import_zod19.z.string().min(1).describe('Business category, niche, keyword, or search term. If the user says "roofers in Denver CO", use query="roofers" and location="Denver, CO". Do not put the location here when it can be separated.'),
+      location: import_zod19.z.string().optional().describe('City, region, country, or service area for the Maps search, e.g. "Denver, CO". Infer from the user request when present.'),
+      gl: import_zod19.z.string().length(2).default("us").describe("Google country code inferred from location."),
+      hl: import_zod19.z.string().length(2).default("en").describe("Language inferred from user request."),
+      maxResults: import_zod19.z.number().int().min(1).max(50).default(10).describe("Number of Google Maps business/profile candidates to return. Default 10. Maximum 50. Use 10 unless the user asks for more.")
+    };
+    NullableString = import_zod19.z.string().nullable();
+    MapsSearchOutputSchema = {
+      query: import_zod19.z.string(),
+      location: import_zod19.z.string().nullable(),
+      searchQuery: import_zod19.z.string(),
+      searchUrl: import_zod19.z.string().url(),
+      extractedAt: import_zod19.z.string(),
+      requestedMaxResults: import_zod19.z.number().int().min(1).max(50),
+      resultCount: import_zod19.z.number().int().min(0).max(50),
+      results: import_zod19.z.array(import_zod19.z.object({
+        position: import_zod19.z.number().int().min(1),
+        name: import_zod19.z.string(),
+        placeUrl: import_zod19.z.string().url(),
+        cid: NullableString,
+        cidDecimal: NullableString,
+        rating: NullableString,
+        reviewCount: NullableString,
+        category: NullableString,
+        address: NullableString,
+        websiteUrl: NullableString,
+        directionsUrl: NullableString,
+        metadata: import_zod19.z.array(import_zod19.z.string())
+      })),
+      durationMs: import_zod19.z.number().int().min(0)
+    };
+    OrganicResultOutput = import_zod19.z.object({
+      position: import_zod19.z.number().int(),
+      title: import_zod19.z.string(),
+      url: import_zod19.z.string(),
+      domain: import_zod19.z.string(),
+      snippet: NullableString
+    });
+    AiOverviewOutput = import_zod19.z.object({
+      detected: import_zod19.z.boolean(),
+      text: NullableString
+    }).nullable();
+    EntityIdsOutput = import_zod19.z.object({
+      kgIds: import_zod19.z.array(import_zod19.z.string()),
+      cids: import_zod19.z.array(import_zod19.z.string()),
+      gcids: import_zod19.z.array(import_zod19.z.string())
+    }).nullable();
+    HarvestPaaOutputSchema = {
+      query: import_zod19.z.string(),
+      location: NullableString,
+      questionCount: import_zod19.z.number().int().min(0),
+      completionStatus: NullableString,
+      questions: import_zod19.z.array(import_zod19.z.object({
+        question: import_zod19.z.string(),
+        answer: NullableString,
+        sourceTitle: NullableString,
+        sourceSite: NullableString
+      })),
+      organicResults: import_zod19.z.array(OrganicResultOutput),
+      aiOverview: AiOverviewOutput,
+      entityIds: EntityIdsOutput,
+      durationMs: import_zod19.z.number().min(0).nullable()
+    };
+    SearchSerpOutputSchema = {
+      query: import_zod19.z.string(),
+      location: NullableString,
+      organicResults: import_zod19.z.array(OrganicResultOutput),
+      localPack: import_zod19.z.array(import_zod19.z.object({
+        position: import_zod19.z.number().int(),
+        name: import_zod19.z.string(),
+        rating: NullableString,
+        reviewCount: NullableString,
+        websiteUrl: NullableString
+      })),
+      aiOverview: AiOverviewOutput,
+      entityIds: EntityIdsOutput
+    };
+    ExtractUrlOutputSchema = {
+      url: import_zod19.z.string(),
+      title: NullableString,
+      headings: import_zod19.z.array(import_zod19.z.object({
+        level: import_zod19.z.number().int(),
+        text: import_zod19.z.string()
+      })),
+      schemaBlockCount: import_zod19.z.number().int().min(0),
+      entityName: NullableString,
+      entityTypes: import_zod19.z.array(import_zod19.z.string()),
+      napScore: import_zod19.z.number().nullable(),
+      missingSchemaFields: import_zod19.z.array(import_zod19.z.string()),
+      screenshotSaved: NullableString
+    };
+    ExtractSiteOutputSchema = {
+      url: import_zod19.z.string(),
+      pageCount: import_zod19.z.number().int().min(0),
+      pages: import_zod19.z.array(import_zod19.z.object({
+        url: import_zod19.z.string(),
+        title: NullableString,
+        schemaTypes: import_zod19.z.array(import_zod19.z.string())
+      })),
+      durationMs: import_zod19.z.number().min(0)
+    };
+    MapsPlaceIntelOutputSchema = {
+      name: import_zod19.z.string(),
+      rating: NullableString,
+      reviewCount: NullableString,
+      category: NullableString,
+      address: NullableString,
+      phone: NullableString,
+      website: NullableString,
+      hoursSummary: NullableString,
+      bookingUrl: NullableString,
+      kgmid: NullableString,
+      cidDecimal: NullableString,
+      cidUrl: NullableString,
+      lat: import_zod19.z.number().nullable(),
+      lng: import_zod19.z.number().nullable(),
+      reviewsStatus: import_zod19.z.string(),
+      reviewsCollected: import_zod19.z.number().int().min(0),
+      reviewTopics: import_zod19.z.array(import_zod19.z.object({
+        label: import_zod19.z.string(),
+        count: import_zod19.z.string()
+      }))
+    };
+    CreditsInfoOutputSchema = {
+      balanceCredits: import_zod19.z.number().nullable(),
+      matchedCost: import_zod19.z.object({
+        label: import_zod19.z.string(),
+        credits: import_zod19.z.number(),
+        unit: import_zod19.z.string(),
+        notes: NullableString
+      }).nullable(),
+      costs: import_zod19.z.array(import_zod19.z.object({
+        key: import_zod19.z.string(),
+        label: import_zod19.z.string(),
+        credits: import_zod19.z.number(),
+        unit: import_zod19.z.string(),
+        notes: NullableString
+      })),
+      ledger: import_zod19.z.array(import_zod19.z.object({
+        createdAt: import_zod19.z.string(),
+        operation: import_zod19.z.string(),
+        credits: import_zod19.z.number(),
+        description: NullableString
+      }))
+    };
+    MapSiteUrlsOutputSchema = {
+      startUrl: import_zod19.z.string(),
+      totalFound: import_zod19.z.number().int().min(0),
+      truncated: import_zod19.z.boolean(),
+      okCount: import_zod19.z.number().int().min(0),
+      redirectCount: import_zod19.z.number().int().min(0),
+      brokenCount: import_zod19.z.number().int().min(0),
+      urls: import_zod19.z.array(import_zod19.z.object({
+        url: import_zod19.z.string(),
+        status: import_zod19.z.number().int().nullable()
+      })),
+      durationMs: import_zod19.z.number().min(0)
+    };
+    YoutubeHarvestOutputSchema = {
+      mode: import_zod19.z.string(),
+      videoCount: import_zod19.z.number().int().min(0),
+      channel: import_zod19.z.object({
+        title: NullableString,
+        subscriberCount: NullableString
+      }).nullable(),
+      videos: import_zod19.z.array(import_zod19.z.object({
+        videoId: import_zod19.z.string(),
+        title: import_zod19.z.string(),
+        channelName: NullableString,
+        views: NullableString,
+        duration: NullableString,
+        url: NullableString
+      }))
+    };
+    FacebookAdSearchOutputSchema = {
+      query: import_zod19.z.string(),
+      advertiserCount: import_zod19.z.number().int().min(0),
+      advertisers: import_zod19.z.array(import_zod19.z.object({
+        name: NullableString,
+        adCount: import_zod19.z.number().int().nullable(),
+        libraryId: NullableString
+      }))
+    };
+    FacebookPageIntelOutputSchema = {
+      advertiserName: NullableString,
+      totalAds: import_zod19.z.number().int().min(0),
+      activeCount: import_zod19.z.number().int().min(0),
+      videoCount: import_zod19.z.number().int().min(0),
+      imageCount: import_zod19.z.number().int().min(0),
+      ads: import_zod19.z.array(import_zod19.z.object({
+        libraryId: NullableString,
+        status: NullableString,
+        creativeType: NullableString,
+        headline: NullableString,
+        cta: NullableString,
+        startDate: NullableString,
+        videoUrl: NullableString,
+        variations: import_zod19.z.number().int().nullable()
+      }))
+    };
+    CreditsInfoInputSchema = {
+      item: import_zod19.z.string().optional().describe('Optional tool, action, or feature to look up, e.g. "maps reviews", "extract_url", or "YouTube transcription"'),
+      includeLedger: import_zod19.z.boolean().default(false).describe("Whether to include recent credit ledger entries")
+    };
+    SearchSerpInputSchema = {
+      query: import_zod19.z.string().min(1).describe('Core search topic only. Separate location when possible. If user says "best dentist in Brooklyn NY serp", use query="best dentist" and location="Brooklyn, NY".'),
+      location: import_zod19.z.string().optional().describe("City, region, or country for geo-targeted results, inferred from user request when present."),
+      gl: import_zod19.z.string().length(2).default("us").describe("Google country code inferred from location or user language."),
+      hl: import_zod19.z.string().default("en").describe("Google interface/content language inferred from user request."),
+      device: import_zod19.z.enum(["desktop", "mobile"]).default("desktop").describe("SERP device context. Use desktop by default; use mobile only when the user asks for mobile rankings."),
+      proxyMode: import_zod19.z.enum(["location", "configured", "none"]).default("location").describe("Proxy targeting mode. Use location by default so city/state searches create or reuse a matching residential proxy. Use configured for the static configured proxy. Use none only for direct-network debugging."),
+      proxyZip: import_zod19.z.string().regex(/^\d{5}$/).optional().describe("Optional US ZIP override for residential location proxy targeting. Use only when the user gives a specific ZIP or city-center proxy targeting needs to be forced."),
+      debug: import_zod19.z.boolean().default(false).describe("Include sanitized browser/session/location diagnostics in the response. Use true when debugging localization, CAPTCHA, or proxy behavior."),
+      pages: import_zod19.z.number().int().min(1).max(2).default(1).describe("Number of result pages to fetch (1\u20132)")
+    };
+    CaptureSerpSnapshotInputSchema = {
+      query: import_zod19.z.string().min(1).describe("Core search query to capture as a structured SERP Intelligence snapshot. Separate the place into location when the user gives a city, region, country, or ZIP."),
+      location: import_zod19.z.string().optional().describe("City, region, country, or service area used for localized Google results. MCP Scraper records location evidence; UULE alone is not proof of localization."),
+      gl: import_zod19.z.string().length(2).default("us").describe("Google country code inferred from the requested market, e.g. us, gb, ca, au."),
+      hl: import_zod19.z.string().default("en").describe("Google interface/content language inferred from the user request."),
+      device: import_zod19.z.enum(["desktop", "mobile"]).default("desktop").describe("SERP device context. Use mobile only when the user asks for mobile rankings or mobile SERP evidence."),
+      proxyMode: import_zod19.z.enum(["location", "configured", "none"]).default("location").describe("Proxy behavior for capture. Use location for localized residential proxy targeting, configured for the static residential proxy, and none only for direct-network debugging."),
+      proxyZip: import_zod19.z.string().regex(/^\d{5}$/).optional().describe("Optional US ZIP override for residential location proxy targeting when a precise city-center or ZIP proxy is needed."),
+      pages: import_zod19.z.number().int().min(1).max(2).default(1).describe("Number of Google result pages to capture. Use 1 normally and 2 only when the user needs deeper ranking evidence."),
+      debug: import_zod19.z.boolean().default(false).describe("Include sanitized browser, proxy, and location diagnostics. Use true when debugging localization, CAPTCHA, proxy selection, or capture reliability."),
+      includePageSnapshots: import_zod19.z.boolean().default(false).describe("Also capture ranking-page snapshots for selected SERP URLs through the same product capture path."),
+      pageSnapshotLimit: import_zod19.z.number().int().min(0).max(10).default(0).describe("Maximum ranking-page snapshots to capture when includePageSnapshots is true. Use 0 when only SERP evidence is needed.")
+    };
+    ScreenshotInputSchema = {
+      url: import_zod19.z.string().url().describe("URL to capture as a full-page screenshot. Use http or https. Pass allowLocal: true to capture localhost or private-network URLs during development."),
+      device: import_zod19.z.enum(["desktop", "mobile"]).default("desktop").describe("Viewport profile. desktop = 1440\xD7900. mobile = 390\xD7844. Use desktop by default; use mobile when the user asks for a mobile view."),
+      allowLocal: import_zod19.z.boolean().default(false).describe("Allow localhost and private-network URLs (127.x, 192.168.x, 10.x, etc.). For local development only \u2014 not for production use.")
+    };
+    CaptureSerpPageSnapshotsInputSchema = {
+      urls: import_zod19.z.array(import_zod19.z.string().url()).min(1).max(25).describe("Public HTTP/HTTPS URLs to capture as SERP Intelligence page snapshots. Do not pass localhost, private IPs, file URLs, or internal admin URLs."),
+      targets: import_zod19.z.array(import_zod19.z.object({
+        url: import_zod19.z.string().url().describe("Public HTTP/HTTPS URL to capture."),
+        sourceKind: import_zod19.z.enum(["organic", "ai_citation", "local_pack_website", "configured_target", "site_subject"]).default("configured_target").describe("Why this page is being captured for SERP Intelligence evidence."),
+        sourcePosition: import_zod19.z.number().int().min(1).optional().describe("Ranking or citation position when the page came from SERP evidence.")
+      }).strict()).min(1).max(25).optional().describe("Structured page snapshot targets. Use this instead of urls when source kind or position should be preserved."),
+      maxConcurrency: import_zod19.z.number().int().min(1).max(5).default(2).describe("Parallel page captures. Use 2 normally; higher values can increase proxy/browser pressure."),
+      timeoutMs: import_zod19.z.number().int().min(1e3).max(6e4).default(15e3).describe("Per-page capture timeout in milliseconds. Increase for slow pages; timeout artifacts are returned as structured capture failures."),
+      debug: import_zod19.z.boolean().default(false).describe("Include sanitized browser/proxy diagnostics for page snapshot debugging. Use true for capture, network, or proxy troubleshooting.")
+    };
+  }
+});
+
 // src/mcp/paa-mcp-server.ts
 function liveWebToolAnnotations(title) {
   return {
@@ -16168,27 +16471,65 @@ function liveWebToolAnnotations(title) {
     openWorldHint: true
   };
 }
+function listSavedReports() {
+  try {
+    const dir = outputBaseDir();
+    return (0, import_node_fs5.readdirSync)(dir).filter((f) => f.endsWith(".md")).map((f) => ({ filename: f, mtimeMs: (0, import_node_fs5.statSync)((0, import_node_path7.join)(dir, f)).mtimeMs })).sort((a, b) => b.mtimeMs - a.mtimeMs).slice(0, 100);
+  } catch {
+    return [];
+  }
+}
+function registerSavedReportResources(server) {
+  server.registerResource(
+    "saved-report",
+    new import_mcp.ResourceTemplate("report://{filename}", {
+      list: () => ({
+        resources: listSavedReports().map((r) => ({
+          uri: `report://${encodeURIComponent(r.filename)}`,
+          name: r.filename,
+          mimeType: "text/markdown"
+        }))
+      })
+    }),
+    {
+      title: "Saved MCP Scraper Reports",
+      description: "Markdown research reports saved by previous MCP Scraper tool calls. Read a report to reuse prior research without re-scraping or spending credits.",
+      mimeType: "text/markdown"
+    },
+    async (uri, variables) => {
+      const requested = Array.isArray(variables.filename) ? variables.filename[0] : variables.filename;
+      const filename = (0, import_node_path7.basename)(decodeURIComponent(String(requested ?? "")));
+      if (!filename.endsWith(".md")) throw new Error("Only saved .md reports can be read");
+      const text = (0, import_node_fs5.readFileSync)((0, import_node_path7.join)(outputBaseDir(), filename), "utf8");
+      return { contents: [{ uri: uri.href, mimeType: "text/markdown", text }] };
+    }
+  );
+}
 function buildPaaExtractorMcpServer(executor, options = {}) {
   const savesReports = options.savesReportsLocally !== false;
   const reportNote = savesReports ? " Saves a full Markdown report locally." : " Reports are returned inline; no files are saved on this hosted endpoint.";
   const withReportNote = (description) => `${description}${reportNote}`;
   const server = new import_mcp.McpServer({ name: "mcp-scraper", version: PACKAGE_VERSION });
+  if (savesReports) registerSavedReportResources(server);
   server.registerTool("harvest_paa", {
     title: "Google PAA + SERP Harvest",
-    description: withReportNote('Best default tool for Google search research. Extracts People Also Ask questions plus answers/source URLs, organic SERP, local pack when present, entity IDs (CID/GCID/KG MID), and AI Overview. Infer the user language: split topic from location (e.g. "best hvac company in Denver CO" => query "best hvac company", location "Denver, CO", gl "us", hl "en"). Use maxQuestions 30 normally, 100-150 for "full", "deep", "all", or comprehensive research. Credits are charged by extracted question; unused request hold is refunded.'),
+    description: withReportNote('Best default tool for Google search research. Extracts People Also Ask questions plus answers/source URLs, organic SERP, local pack when present, entity IDs (CID/GCID/KG MID), and AI Overview. Infer the user language: split topic from location (e.g. "best hvac company in Denver CO" => query "best hvac company", location "Denver, CO", gl "us", hl "en"). Use maxQuestions 30 normally, 100-200 for "full", "deep", "all", or comprehensive research. Deep harvests above 100 questions can run for several minutes with no interim progress \u2014 warn the user before starting one and keep maxQuestions at or below 100 unless they explicitly want a deep harvest. Credits are charged by extracted question; unused request hold is refunded.'),
     inputSchema: HarvestPaaInputSchema,
+    outputSchema: HarvestPaaOutputSchema,
     annotations: liveWebToolAnnotations("Google PAA + SERP Harvest")
   }, async (input) => formatHarvestPaa(await executor.harvestPaa(input), input));
   server.registerTool("search_serp", {
     title: "Google SERP Lookup",
     description: withReportNote("Fast Google SERP lookup without PAA expansion. Use when the user asks for rankings, organic results, local pack, quick SERP, or positions. Split topic from location and infer gl/hl from the user request."),
     inputSchema: SearchSerpInputSchema,
+    outputSchema: SearchSerpOutputSchema,
     annotations: liveWebToolAnnotations("Google SERP Lookup")
   }, async (input) => formatSearchSerp(await executor.searchSerp(input), input));
   server.registerTool("extract_url", {
     title: "Single URL Extract",
     description: withReportNote("Extract structured data from one public URL: page content as Markdown, heading structure, JSON-LD schema, entity details, NAP score, metadata, and missing schema fields. Use when the user provides a single URL or asks to inspect/scrape one page."),
     inputSchema: ExtractUrlInputSchema,
+    outputSchema: ExtractUrlOutputSchema,
     annotations: liveWebToolAnnotations("Single URL Extract")
   }, async (input) => formatExtractUrl(await executor.extractUrl(input), input));
   server.registerTool("map_site_urls", {
@@ -16202,6 +16543,7 @@ function buildPaaExtractorMcpServer(executor, options = {}) {
     title: "Multi-Page Site Extract",
     description: withReportNote("Run multi-page extraction across a public website. Returns per-page titles, H1s, metadata, headings, schema/entity data, canonical URLs, and content. Use for website audits, competitor audits, and full-site extraction."),
     inputSchema: ExtractSiteInputSchema,
+    outputSchema: ExtractSiteOutputSchema,
     annotations: liveWebToolAnnotations("Multi-Page Site Extract")
   }, async (input) => formatExtractSite(await executor.extractSite(input), input));
   server.registerTool("youtube_harvest", {
@@ -16241,6 +16583,7 @@ function buildPaaExtractorMcpServer(executor, options = {}) {
     title: "Google Maps Business Profile Details",
     description: withReportNote('Extract Google Maps business intelligence for one known/named business: rating, review count, category, address, phone, website, hours, booking URL, review histogram, review topics, about attributes, entity IDs, and optional review cards. Do not use this for category searches, local market prospect lists, or requests for multiple GMB/GBP profiles; use maps_search first for those. Split business name from location (e.g. "Elite Roofing Denver CO" => businessName "Elite Roofing", location "Denver, CO"). Pass includeReviews true when the user asks for reviews/customer pain.'),
     inputSchema: MapsPlaceIntelInputSchema,
+    outputSchema: MapsPlaceIntelOutputSchema,
     annotations: liveWebToolAnnotations("Google Maps Business Profile Details")
   }, async (input) => formatMapsPlaceIntel(await executor.mapsPlaceIntel(input), input));
   server.registerTool("maps_search", {
@@ -16254,6 +16597,7 @@ function buildPaaExtractorMcpServer(executor, options = {}) {
     title: "MCP Scraper Credits & Costs",
     description: "Answer questions about MCP Scraper credits: current credit balance, what a specific tool/action costs, the full cost table, and optionally recent credit ledger entries. Does not expose payment methods or credit card information.",
     inputSchema: CreditsInfoInputSchema,
+    outputSchema: CreditsInfoOutputSchema,
     annotations: {
       title: "MCP Scraper Credits & Costs",
       readOnlyHint: true,
@@ -16264,12 +16608,15 @@ function buildPaaExtractorMcpServer(executor, options = {}) {
   }, async (input) => formatCreditsInfo(await executor.creditsInfo(input), input));
   return server;
 }
-var import_mcp;
+var import_mcp, import_node_fs5, import_node_path7;
 var init_paa_mcp_server = __esm({
   "src/mcp/paa-mcp-server.ts"() {
     "use strict";
     import_mcp = require("@modelcontextprotocol/sdk/server/mcp.js");
+    import_node_fs5 = require("fs");
+    import_node_path7 = require("path");
     init_version();
+    init_mcp_response_formatter();
     init_mcp_tool_schemas();
     init_mcp_response_formatter();
   }
@@ -16397,7 +16744,10 @@ function mcpAuthError() {
   });
   return new Response(body, {
     status: 401,
-    headers: { "Content-Type": "application/json" }
+    headers: {
+      "Content-Type": "application/json",
+      "WWW-Authenticate": 'Bearer realm="mcp-scraper", error="invalid_token", error_description="Pass an MCP Scraper API key as x-api-key or Bearer token"'
+    }
   });
 }
 async function requireMcpCallerKey(c) {
@@ -16833,7 +17183,7 @@ async function processJob(job) {
     const opts = typeof job.options === "string" ? JSON.parse(job.options) : job.options;
     const result = await harvest({
       ...opts,
-      kernelApiKey: process.env.KERNEL_API_KEY,
+      kernelApiKey: browserServiceApiKey(),
       headless: true,
       format: "json",
       outputDir: "/tmp/paa-output-api",
@@ -16898,6 +17248,7 @@ var init_worker = __esm({
   "src/api/worker.ts"() {
     "use strict";
     init_db();
+    init_browser_service_env();
     init_harvest();
     init_webhook();
     init_rates();
@@ -17000,6 +17351,8 @@ var init_server = __esm({
   "src/api/server.ts"() {
     "use strict";
     init_harvest_timeout();
+    init_browser_service_env();
+    init_outbound_sanitize();
     init_registry();
     init_template();
     init_og();
@@ -17316,7 +17669,7 @@ var init_server = __esm({
       try {
         const result = await harvest({
           ...options,
-          kernelApiKey: process.env.KERNEL_API_KEY?.trim(),
+          kernelApiKey: browserServiceApiKey(),
           headless: true,
           format: "json",
           outputDir: "/tmp/paa-output-api",
@@ -17331,7 +17684,7 @@ var init_server = __esm({
           if (diff > 0) await creditMc(user.id, diff, LedgerOperation.PAA_REFUND, "overestimate refund");
           else if (diff < 0) await debitMc(user.id, -diff, LedgerOperation.PAA, options.query);
         }
-        return c.json({ job_id: jobId, status: "done", result, attempts });
+        return c.json({ job_id: jobId, status: "done", result: sanitizeHarvestResult(result), attempts: sanitizeAttempts(attempts) });
       } catch (err) {
         const problem = classifyHarvestProblem(err);
         const response = harvestProblemResponse(problem);
@@ -17339,18 +17692,19 @@ var init_server = __esm({
         if (problem.terminalStatus === "cancelled" || c.req.raw.signal.aborted) {
           await cancelJob(jobId, serializeHarvestProblem(problem));
           await creditMc(user.id, syncCost, LedgerOperation.REFUND, "cancelled call");
-          return c.json({ job_id: jobId, status: "cancelled", ...response, attempts }, problem.httpStatus);
+          return c.json({ job_id: jobId, status: "cancelled", ...response, attempts: sanitizeAttempts(attempts) }, problem.httpStatus);
         }
         await failJob(jobId, serializeHarvestProblem(problem));
         await creditMc(user.id, syncCost, LedgerOperation.REFUND, "failed call");
-        return c.json({ job_id: jobId, status: "failed", ...response, attempts }, problem.httpStatus);
+        return c.json({ job_id: jobId, status: "failed", ...response, attempts: sanitizeAttempts(attempts) }, problem.httpStatus);
       }
     });
     app.get("/jobs/:id", auth, async (c) => {
       const job = await getJob(c.req.param("id"), c.get("user").id);
       if (!job) return c.json({ error: "Job not found" }, 404);
       const attempts = await listHarvestAttempts(job.id, c.get("user").id);
-      return c.json({ ...job, attempts });
+      const safeResult = job.result && typeof job.result === "object" ? sanitizeHarvestResult(job.result) : job.result;
+      return c.json({ ...job, result: safeResult, attempts: sanitizeAttempts(attempts) });
     });
     app.get("/jobs", auth, async (c) => {
       return c.json(await listJobs(c.get("user").id));
@@ -17449,7 +17803,7 @@ var init_server = __esm({
       const { ok: euOk, balance_mc: euBal } = await debitMc(user.id, MC_COSTS.page_scrape, LedgerOperation.EXTRACT_URL, new URL(canonicalUrl).hostname);
       if (!euOk) return c.json(insufficientBalanceResponse(euBal, MC_COSTS.page_scrape), 402);
       try {
-        const kernelApiKey = process.env.KERNEL_API_KEY?.trim();
+        const kernelApiKey = browserServiceApiKey();
         const device = screenshotDevice === "mobile" ? "mobile" : "desktop";
         const [result, pageData] = await Promise.all([
           extractKpo({ url: canonicalUrl, kernelApiKey }),
@@ -17487,7 +17841,7 @@ var init_server = __esm({
           startUrl: parsed.href,
           maxUrls: Math.min(2e3, Math.max(1, body.maxUrls ?? 500)),
           concurrency: Math.min(20, Math.max(1, body.concurrency ?? 12)),
-          kernelApiKey: body.browserFallback ?? body.kernelFallback ? process.env.KERNEL_API_KEY : void 0
+          kernelApiKey: body.browserFallback ?? body.kernelFallback ? browserServiceApiKey() : void 0
         });
         await logRequestEvent({
           userId: user.id,
@@ -17527,7 +17881,7 @@ var init_server = __esm({
         const result = await extractSite({
           startUrl: parsed.href,
           maxPages: Math.min(200, Math.max(1, body.maxPages ?? 100)),
-          kernelApiKey: body.browserFallback ?? body.kernelFallback ? process.env.KERNEL_API_KEY : void 0
+          kernelApiKey: body.browserFallback ?? body.kernelFallback ? browserServiceApiKey() : void 0
         });
         const pageCount = result.pages?.length ?? 1;
         const actualSiteMc = pageCount * MC_COSTS.page_scrape;
@@ -17797,10 +18151,10 @@ var init_server = __esm({
 });
 
 // bin/api-server.ts
-var import_node_fs5 = require("fs");
+var import_node_fs6 = require("fs");
 function loadDotEnv() {
   try {
-    for (const line of (0, import_node_fs5.readFileSync)(".env", "utf8").split("\n")) {
+    for (const line of (0, import_node_fs6.readFileSync)(".env", "utf8").split("\n")) {
       const eq = line.indexOf("=");
       if (eq < 1 || line.trimStart().startsWith("#")) continue;
       const k = line.slice(0, eq).trim();