mcp-scraper 0.1.7 → 0.1.8

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");
   }
 });
@@ -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
@@ -15199,12 +15276,12 @@ var PACKAGE_VERSION;
 var init_version = __esm({
   "src/version.ts"() {
     "use strict";
-    PACKAGE_VERSION = "0.1.7";
+    PACKAGE_VERSION = "0.1.8";
   }
 });
 
 // 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 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";
@@ -15301,6 +15378,120 @@ var init_mcp_tool_schemas = __esm({
       })),
       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),
@@ -15505,7 +15696,7 @@ 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 +15722,14 @@ 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 +15776,31 @@ ${serpRows}` : "";
   const full = `# PAA Report: "${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 +15838,29 @@ ${localRows}` : "";
   const full = `# SERP Report: "${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 +15929,27 @@ ${bodyMd.slice(0, 3e3)}${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 +16019,19 @@ ${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 +16228,26 @@ ${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 +16396,28 @@ ${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);
@@ -16177,18 +16480,21 @@ function buildPaaExtractorMcpServer(executor, options = {}) {
     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.'),
     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 +16508,7 @@ 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 +16548,7 @@ 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 +16562,7 @@ 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,
@@ -16833,7 +17142,7 @@ 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 +17207,7 @@ var init_worker = __esm({
   "src/api/worker.ts"() {
     "use strict";
     init_db();
+    init_browser_service_env();
     init_harvest();
     init_webhook();
     init_rates();
@@ -17000,6 +17310,8 @@ 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 +17628,7 @@ 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 +17643,7 @@ 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 +17651,19 @@ 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 +17762,7 @@ 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 +17800,7 @@ 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 +17840,7 @@ 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;