跳到正文
DEV Community· Kiranvia·· 4 天前AI 評分52

Kiranvia 如何在 n8n 建立無需瀏覽器的 B2B 公司資料豐富化流程,取代 5 個串接 API

How I Built a Zero-Browser B2B Company Enrichment Pipeline in n8n (Replacing 5 Chained APIs)sda

AI 導讀

Kiranvia 分享在 n8n 建立無需瀏覽器的 B2B 公司資料豐富化流程,以單次 HTTP 呼叫取代 4–5 個串接 API。流程會為每個網域輸出結構化資料及 buyingSignalScore(0–100),並以串流 HTTP、Cheerio 和 Node.js 內置功能檢查 DNS、TLS 及最多 5 個高訊號內頁,在 256 MB 容器中峯值記憶體約 160 MB。

正文

Kiranvia

If you build outbound pipelines in n8n, Clay, or Make, you probably know the pain of enriching a list of company domains:

  1. Call one tool for emails & socials
  2. Call another for tech stack detection
  3. Call another for DNS / SPF / DMARC & SSL health
  4. Call another to check careers/ATS pages and /pricing
  5. Stitch 4–5 mismatched JSON payloads together and pay 4–5x the cost

I wanted one lightweight HTTP call that runs without headless browsers or proxies and returns a single flat + structured JSON row per domain — complete with a deterministic buyingSignalScore (0–100) and human-readable outreach hooks.


The Architecture (Zero Browser, 256 MB RAM)

Instead of launching Playwright/Chromium (slow cold starts + heavy memory), the enrichment runs on pure streaming HTTP + Cheerio + Node.js built-ins (dns/promises, node:tls):

  1. Parallel base checks: Fetch homepage HTML (byte-capped stream), live DNS (MX, SPF, DMARC), and live port-443 TLS certificate simultaneously.
  2. Smart subpage discovery: Auto-discovers and inspects up to 5 high-signal internal pages (/pricing, /careers, /contact, /blog, /security) — parsing one DOM at a time so peak memory stays around ~160 MB on a 256 MB container.
  3. Hidden tech stack trick via SPF records: Modern Next.js/Astro sites often hide marketing scripts behind proxies, so HTML alone misses tools. Parsing the company's live DNS SPF TXT record (include:...hubspotemail.net, spf.mtasv.net, mail.zendesk.com, sendgrid.net, amazonses.com) reliably surfaces CRM, email, and support platforms with zero HTML noise.
  4. Cloudflare email decoding: Decodes /cdn-cgi/l/email-protection#... hex XOR links and strips junk/placeholder addresses (wixpress, sentry, yourcompany.com, @2x.png).
  5. AI Search Readiness (GEO): Audits /llms.txt (filtering out SPA HTML 200 fallbacks), robots.txt rules for GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and /sitemap.xml <lastmod> freshness.

Real Sample Output (linear.app)

{
  "domain": "linear.app",
  "companyName": "Linear",
  "buyingSignalScore": 77,
  "buyingSignalTier": "hot",
  "buyingSignals": [
    "Actively hiring (Ashby ATS detected)",
    "Clear freemium pricing motion ($0, $10)",
    "Enterprise compliance (SOC 2, ISO 27001, GDPR)",
    "Active content velocity (updated 0d ago)"
  ],
  "primaryEmail": "hello@linear.app",
  "emailProvider": "Google Workspace",
  "techStackSummary": ["Next.js", "HubSpot", "Postmark", "Cloudflare", "AWS SES"],
  "atsPlatform": "Ashby",
  "pricingModel": "freemium",
  "hasSsl": true,
  "hasSpf": true,
  "hasDmarc": true,
  "hasLlmsTxt": true,
  "seoScore": 82
}

Free n8n Workflow Template (Copy & Import)

I put together a ready-to-import n8n workflow that:

Also Works with Claude / Cursor via MCP

If you use the Apify MCP server (https://mcp.apify.com), AI agents can call kiranvia/b2b-company-intelligence directly to research a prospect domain before drafting outreach.

Happy to hear feedback on any additional ATS platforms, tech signatures, or buying signals you'd find useful in your outbound workflows!

來源:DEV Community · dev.to