#!/usr/bin/env bun /** * Crawl a manufacturer's website and upsert their products into the GearBox catalog. * * Usage: * bun run scripts/crawl-manufacturer.ts --manufacturer=apidura * bun run scripts/crawl-manufacturer.ts --manufacturer=canyon --dry-run * * Env vars required: * ANTHROPIC_API_KEY — Anthropic API key * GEARBOX_URL — Base URL of the GearBox instance (default: http://localhost:3000) * GEARBOX_API_KEY — GearBox API key with write access */ import Anthropic from "@anthropic-ai/sdk"; import { CATEGORIES } from "./taxonomy/categories.ts"; import { TAGS } from "./taxonomy/tags.ts"; const GEARBOX_URL = process.env.GEARBOX_URL ?? "http://localhost:3000"; const GEARBOX_API_KEY = process.env.GEARBOX_API_KEY ?? ""; const ANTHROPIC_API_KEY = process.env.ANTHROPIC_API_KEY ?? ""; const MODEL = "claude-haiku-4-5-20251001"; const MAX_TOOL_ROUNDS = 30; // safety limit // ── Parse CLI args ──────────────────────────────────────────────── const args = Object.fromEntries( process.argv .slice(2) .filter((a) => a.startsWith("--")) .map((a) => { const [k, v] = a.slice(2).split("="); return [k, v ?? "true"]; }), ); const manufacturerSlug = args["manufacturer"]; const dryRun = args["dry-run"] === "true"; if (!manufacturerSlug) { console.error("Usage: bun run scripts/crawl-manufacturer.ts --manufacturer="); process.exit(1); } if (!GEARBOX_API_KEY) { console.error("GEARBOX_API_KEY env var is required"); process.exit(1); } if (!ANTHROPIC_API_KEY) { console.error("ANTHROPIC_API_KEY env var is required"); process.exit(1); } // ── Fetch manufacturer from GearBox ────────────────────────────── async function fetchManufacturer(slug: string) { const res = await fetch(`${GEARBOX_URL}/api/manufacturers/${slug}`); if (!res.ok) { throw new Error(`Manufacturer not found: ${slug} (HTTP ${res.status})`); } return res.json() as Promise<{ id: number; name: string; slug: string; website: string; tier: number; country: string | null; }>; } // ── Tool: fetch a web page ──────────────────────────────────────── async function fetchPage(url: string): Promise { try { const res = await fetch(url, { headers: { "User-Agent": "Mozilla/5.0 (compatible; GearBox-Catalog-Bot/1.0)", Accept: "text/html,application/xhtml+xml", }, signal: AbortSignal.timeout(15_000), }); if (!res.ok) return `HTTP ${res.status} for ${url}`; const html = await res.text(); // Strip scripts, styles, and excessive whitespace for token efficiency return html .replace(/]*>[\s\S]*?<\/script>/gi, "") .replace(/]*>[\s\S]*?<\/style>/gi, "") .replace(//g, "") .replace(/\s{3,}/g, " ") .slice(0, 60_000); // cap at 60k chars to stay within context } catch (err) { return `Error fetching ${url}: ${(err as Error).message}`; } } // ── Build system prompt ─────────────────────────────────────────── function buildSystemPrompt(manufacturer: Awaited>) { return `You are a product data extraction agent for GearBox, a gear management app for bikepacking, cycling, and hiking. Your task: crawl ${manufacturer.name}'s website (${manufacturer.website}) and extract their complete product catalog. For each product, extract: - model: string (product name WITHOUT the brand prefix) - category: one of [${CATEGORIES.join(", ")}] - weightGrams: number | null (weight in grams — convert if shown in oz/lbs/kg) - priceCents: number | null (MSRP in cents, base currency) - priceCurrency: string (ISO currency code — "EUR" for DE brands, "USD" for US, "GBP" for GB, etc.) - description: string | null (1-3 sentence product description) - sourceUrl: string (direct product page URL) - tags: string[] (from this list only: [${TAGS.join(", ")}]) Rules: - model must NOT include the brand name (e.g., "Terrapin System" not "Revelate Designs Terrapin System") - Only include outdoor/adventure/cycling products. Skip accessories under €5, clothing if not relevant to the target categories. - If weight is not listed on a product page, use null — do not guess. - Assign 2-5 relevant tags per item. - Extract every product in their catalog, not just featured ones. Navigate to all relevant subcategories. When done, output a JSON array of product objects as your final message. Do not wrap in markdown — raw JSON only. Example output: [ { "model": "Expedition Handlebar Pack", "category": "bags", "weightGrams": 300, "priceCents": 16000, "priceCurrency": "GBP", "description": "14L waterproof handlebar roll bag with internal dry bag and accessory pocket.", "sourceUrl": "https://apidura.com/shop/expedition-handlebar-pack/", "tags": ["bikepacking", "handlebar-bag", "bike-bag"] } ]`; } // ── Agentic tool-use loop ───────────────────────────────────────── type CatalogItem = { model: string; category: string; weightGrams: number | null; priceCents: number | null; priceCurrency: string; description: string | null; sourceUrl: string; tags: string[]; }; async function runCrawlAgent(manufacturer: Awaited>): Promise { const client = new Anthropic({ apiKey: ANTHROPIC_API_KEY }); const tools: Anthropic.Tool[] = [ { name: "fetch_page", description: "Fetch the HTML content of a URL. Use this to explore the manufacturer's website and product pages.", input_schema: { type: "object" as const, properties: { url: { type: "string", description: "The URL to fetch" }, }, required: ["url"], }, }, ]; const messages: Anthropic.MessageParam[] = [ { role: "user", content: `Crawl ${manufacturer.name}'s website at ${manufacturer.website} and extract their complete product catalog. Start with the homepage or sitemap, navigate to all product categories, and return the full product list as JSON.`, }, ]; let rounds = 0; while (rounds < MAX_TOOL_ROUNDS) { rounds++; console.log(` [round ${rounds}] calling model...`); const response = await client.messages.create({ model: MODEL, max_tokens: 8192, system: buildSystemPrompt(manufacturer), tools, messages, }); // Add assistant response to history messages.push({ role: "assistant", content: response.content }); if (response.stop_reason === "end_turn") { // Final message — extract JSON from text content const textBlock = response.content.find((b) => b.type === "text"); if (!textBlock || textBlock.type !== "text") { throw new Error("Agent finished without text output"); } return parseAgentOutput(textBlock.text); } if (response.stop_reason !== "tool_use") { throw new Error(`Unexpected stop reason: ${response.stop_reason}`); } // Process tool calls const toolResults: Anthropic.ToolResultBlockParam[] = []; for (const block of response.content) { if (block.type !== "tool_use") continue; if (block.name === "fetch_page") { const { url } = block.input as { url: string }; console.log(` [tool] fetch_page ${url}`); const content = await fetchPage(url); toolResults.push({ type: "tool_result", tool_use_id: block.id, content, }); } } messages.push({ role: "user", content: toolResults }); } throw new Error(`Agent exceeded ${MAX_TOOL_ROUNDS} tool rounds without finishing`); } function parseAgentOutput(text: string): CatalogItem[] { // Handle agent wrapping output in markdown code blocks const cleaned = text.replace(/^```json\s*/i, "").replace(/\s*```$/i, "").trim(); const parsed = JSON.parse(cleaned); if (!Array.isArray(parsed)) throw new Error("Agent output is not a JSON array"); return parsed; } // ── Upsert to GearBox API ───────────────────────────────────────── async function upsertItems( slug: string, items: CatalogItem[], ): Promise<{ created: number; updated: number }> { const payload = items.map((item) => ({ manufacturerSlug: slug, model: item.model, category: item.category, weightGrams: item.weightGrams ?? undefined, priceCents: item.priceCents ?? undefined, description: item.description ?? undefined, sourceUrl: item.sourceUrl, tags: item.tags, })); // Chunk into batches of 100 (API limit) let totalCreated = 0; let totalUpdated = 0; for (let i = 0; i < payload.length; i += 100) { const batch = payload.slice(i, i + 100); const res = await fetch(`${GEARBOX_URL}/api/global-items/bulk`, { method: "POST", headers: { "Content-Type": "application/json", "X-API-Key": GEARBOX_API_KEY, }, body: JSON.stringify({ items: batch }), }); if (!res.ok) { const err = await res.text(); throw new Error(`Bulk upsert failed (HTTP ${res.status}): ${err}`); } const result = await res.json() as { created: number; updated: number }; totalCreated += result.created; totalUpdated += result.updated; console.log(` batch ${Math.floor(i / 100) + 1}: +${result.created} new, ~${result.updated} updated`); } return { created: totalCreated, updated: totalUpdated }; } // ── Main ────────────────────────────────────────────────────────── async function main() { console.log(`\nCrawling manufacturer: ${manufacturerSlug}`); if (dryRun) console.log("DRY RUN — products will not be saved\n"); const manufacturer = await fetchManufacturer(manufacturerSlug); console.log(`Found: ${manufacturer.name} (${manufacturer.website})\n`); console.log("Starting agent crawl..."); const items = await runCrawlAgent(manufacturer); console.log(`\nAgent extracted ${items.length} products`); if (dryRun) { console.log("\nDry run output (first 3 items):"); console.log(JSON.stringify(items.slice(0, 3), null, 2)); return; } console.log("\nUpserting to catalog..."); const { created, updated } = await upsertItems(manufacturerSlug, items); console.log(`\nDone: ${created} created, ${updated} updated`); } main().catch((err) => { console.error(err); process.exit(1); });