Research web topics and email a Claude report via Gmail using SerpApi, Jina.ai and Firecrawl

Submit a research topic through a form and receive a professionally styled executive report in your inbox — fully automated, with built-in scraping resilience.

The workflow searches Google via SerpApi, scrapes each result with Jina.ai (free, no key needed), and uses Claude to extract key findings. If a page is blocked by a CAPTCHA or login wall, it automatically retries with Firecrawl. Blocked sources are gracefully skipped after two attempts. Once all sources are processed, Claude synthesises a structured executive report and delivers it as a styled HTML email via Gmail.

How it works

1. A web form collects the research topic, number of sources (5–7), and recipient email
2. SerpApi searches Google and returns a buffer of results (2× requested + 3 to survive domain filtering)
3. Junk domains are filtered out automatically (Reddit, YouTube, Twitter, PDFs, etc.)
4. Each URL is processed one at a time in a serial loop:

• Round 1 — Jina.ai: free Markdown scraper, no API key required
• Claude checks the content — if it's a CAPTCHA or wall, it returns RETRY_NEEDED
• Round 2 — Firecrawl: paid fallback scraper retries the blocked URL
• If still blocked, the source is marked as unavailable and the loop continues

1. All extracted findings are aggregated and Claude writes a structured executive report (Executive Summary, Key Findings, Detailed Analysis, Data & Evidence, Conclusions, Sources)
2. The report is converted to styled HTML (with tables, headings, and lists) and emailed to the recipient

Setup steps

Required credentials

Error handler (optional)

The workflow includes a built-in error handler that captures the failed node name, error message, and execution URL. To activate it: Workflow Settings → Error Workflow → select this workflow. Add a Slack or Gmail node after Format Error to receive failure alerts.

Nodes used

• n8n Form Trigger — collects topic, source count, and recipient email
• HTTP Request × 3 — SerpApi (Google Search), Jina.ai (primary scraper), Firecrawl (fallback scraper)
• Code × 6 — URL filtering, response normalisation, prompt assembly, HTML rendering
• Split In Batches — serial loop (one URL at a time, prevents rate limit collisions)
• IF × 2 — CAPTCHA/block detection after each scrape attempt
• Wait — 3-second pause before Firecrawl retry
• Basic LLM Chain × 3 — page analysis (×2) and report synthesis (×1), all powered by Claude
• Aggregate — collects all per-URL findings before synthesis
• Gmail — sends the final HTML report
• Error Trigger + Set — error handler sub-flow

Notes

• Jina.ai is free and works without an API key for most public pages
• Firecrawl is only called when Jina is blocked — most runs won't consume Firecrawl credits
• SerpApi fetches numSources × 2 + 3 results to ensure enough survive domain filtering
• Claude model is set to claude-sonnet-4-5 — swap to any Anthropic model in the three Claude nodes
• The HTML email renders markdown tables, headings, lists, and bold correctly in Gmail