Web Search & Page Scraper
One query in, whole pages out. The agent runs your search, collects the result links, then fetches each page and extracts the content — title, description, headings, body text, contact details and structured data — as one clean table you can export or hand to an AI agent.
- Type a search query — the agent gets the ranked result links and then scrapes every one of those pages.
- Each page comes back with its title, meta, headings, readable body text, emails, phones, socials and JSON-LD.
- Google first: with a Google API key (or a residential proxy) you get real google.com rankings; otherwise results come from a clearly labelled fallback engine.
- Built for agents too — same run over the REST API and the MCP server, so Hermes, Claude, n8n or Zapier can search and read the web in one call.
Last updated: June 2026
Built for operators, not for clicking.
Links aren't the answer
Every result URL is fetched and parsed in the same run — title, meta, headings, readable body text and word count — so you get what the pages actually say, not ten blue links.
Honest about the source
Serper, SerpAPI, Google Custom Search or a direct google.com fetch when one is available — and a clearly labelled fallback engine when it isn't. Every row carries the engine that produced it.
More than text
JSON-LD schema, Open Graph and Twitter cards, canonical URLs, publish dates, authors, on-page links, images, emails, phone numbers and social profiles — extracted per page.
REST + MCP
One inline REST call for a single look-up, a background job for big sweeps, and three MCP tools so Hermes, Claude, n8n or Zapier can search and read the web directly.
Recent runs across the platform.
Web Search & Page Scraper questions.
What exactly does one run return?
Two layers. First the ranked search results — position, title, URL, domain and snippet. Then, for every link, the scraped page itself: final URL after redirects, HTTP status, title, meta description, canonical, language, H1/H2 headings, readable body text, word count, emails, phone numbers, social profiles, Open Graph / Twitter cards and any JSON-LD structured data.
Does it really use Google?
It tries Google first through a provider chain — Serper, SerpAPI, Google's official Custom Search JSON API, then a direct google.com fetch (which works through a residential proxy). google.com refuses plain datacenter IPs with a JavaScript / CAPTCHA wall, so without a Google provider the run falls back to another engine and labels every row with the engine that produced it. A fallback is never passed off as a Google ranking.
Can I scrape specific URLs without searching?
Yes — leave the query empty and paste URLs, one per line. The agent skips the search step and scrapes exactly those pages with the same extraction.
Can my AI agent call this?
POST /api/search returns the whole payload inline, POST /api/tool-run runs it as a background job, and the ConvertFleet MCP server exposes it as google_search, google_search_scrape and scrape_web_pages.
How many results and pages do I get per run?
Web search has its own per-plan budget, because one run is both N result links and N full page fetches. Starter: 10 results + 10 pages, 2 queries, 5,000 chars of text per page. Pro: 50 + 50, 10 queries, 50,000 chars. Gold: 200 + 200, 25 queries, full-page text, widest parallel fetching and a 30-minute window. Queries in one run share the ceiling rather than multiplying it. See pricing.
What happens when a result page blocks the scraper?
That row comes back with the reason in its status column (HTTP 403, timeout, non-HTML…) while the rest of the run continues. If every page refuses, the run says so in its warnings; if no engine answers at all, the run fails honestly instead of returning an empty green result.
Run your first Web Search & Page Scraper agent.
First 20 records free. No card.