Website Contact Finder: emails, phones and socials from any site

Contact scrapers in this store tend to fail in the same three ways, and the reviews say it better than any marketing page could:

1. The email is right there on the site and the scraper does not retrieve it. The most repeated complaint in this category, word for word.
2. You pay per lead, so you cannot predict the bill. The advertised price is per page, the real one is per contact - and a list of 500 sites can cost ten times what you planned, or nothing at all, depending on luck.
3. You get a value with no source, and sometimes a guess dressed up as verified. If an email was reconstructed from a pattern, calling it "verified" is not a detail - it is the difference between a working campaign and a bounced one.

What this Actor does instead

• Crawls the pages where contacts actually live. The address you give, plus every contact / about / imprint / support / privacy page linked from it (up to maxPagesPerSite, default 5). More pages, more real emails - and no browser, no proxies.
• Every value carries its source. Each email and phone has foundOn - the exact page it was read from, and kind - whether it came from a mailto:/tel: link or from the page text. Nothing is generated, nothing is guessed, no pattern is filled in.
• Pay only for sites that delivered. A site that returns at least one email or phone costs one event. A site with no contacts is free - you still get its row, with the reason. A row served from the cache is free too. That is the whole pricing model, no tiers to decode.
• The same site sent twice is crawled and billed once. apify.com, https://apify.com, https://www.apify.com/ and simplify.hr/contact are one site, not four - duplicates are dropped by host before any request, and SUMMARY.duplicatesRemoved tells you how many. Different subdomains stay different sites.
• Nothing disappears silently. Every unique input gets a row and one of six statuses: OK (contacts found), NO_CONTACTS (with how many pages were checked), BLOCKED (the site answered with an anti-bot challenge page - free, HTTP status in reason), UNREACHABLE (the HTTP code or the exact network error in reason), INVALID_URL (the line is not a http(s) address - free) or ERROR (that one line failed unexpectedly and the rest of the run continued). Compare that with a run that returns fewer rows than you sent.

What you get

One row per website:

{
  "website": "apify.com",
  "url": "https://apify.com",
  "domain": "apify.com",
  "company": "Apify",
  "status": "OK",
  "emails": [
    {"value": "sales@apify.com", "foundOn": "https://apify.com/contact", "kind": "mailto", "role": true}
  ],
  "phones": [
    {"value": "+420 123 456 789", "e164": "+420123456789", "foundOn": "https://apify.com/contact", "kind": "tel-link"}
  ],
  "socials": {"linkedin": "https://linkedin.com/company/apify", "github": "https://github.com/apify"},
  "address": {"street": "Vodickova 704/36", "city": "Praha", "postalCode": "11000", "country": "CZ"},
  "pagesScanned": ["https://apify.com", "https://apify.com/contact"],
  "pagesFailed": [],
  "reason": "",
  "fromCache": false,
  "scannedAt": "2026-11-05T09:12:44+00:00"
}

Social profiles collected when linked on the site: LinkedIn, X, Facebook, Instagram, YouTube, GitHub, TikTok, Telegram, Threads. Address comes from the site's own schema.org (JSON-LD) markup and is null when the site does not publish one.

Price

Per site that returned contacts - not per page, not per address, not per email inside the row. A site with nothing to give costs $0.00, and the row still tells you what was checked. Long lists behave predictably: 500 sites that all have contacts cost 500 events, not 1,400. Duplicates are dropped before crawling, and a cached row is not charged at all.

You set the cap and the platform enforces it. Every run has a hard spending limit (maxTotalChargeUsd - in the Console or in the API call). We lowered the smallest limit the Actor accepts to $0.02, the price of one result, so a run can never cost more than the number you chose. No tiers to decode, no invoice surprises.

For a long list, set the cap deliberately. 500 sites that all have contacts cost $10, while the platform default is about $4.8 - if your list is that long, raise the cap before the run. If the cap is reached, the SDK documentation says the platform stops that run. In our own two tests, with a $0.02 and a $0.04 cap, the run finished with every row intact and nothing was charged beyond the cap - we have not reproduced an abort ourselves, so treat it as the documented behaviour and not as our promise. Either way SUMMARY is saved as the run goes (every 25 sites, partial: true), so rows already found are not lost, and SUMMARY.cappedNotCharged counts the sites that were reported but not billed.

Honest limits

• HTTP only, no JavaScript. Sites that draw their contact block with client-side JS will come back as NO_CONTACTS. We say so instead of pretending. A browser-based mode is not included - it would multiply the price of every run.

• No deliverability claims. We do not say an email is "verified", "valid" or "deliverable", because we did not run an SMTP handshake. We say where we found it.

• A phone is returned exactly as it is written on the page. 8 (495) 123-45-67 stays exactly that. We never glue a + in front of a number that did not have one - that turned a Moscow number into a Vietnamese one in an earlier build. e164 appears only when the page itself wrote the international form.

• A bare number is not a phone. A number written without + is taken only when a phone word (phone, tel, call, contact…) stands next to it, or when it is a tel: link; a + number is taken as written. That is how a code sample on python.org - the Fibonacci sequence 8 13 21 34 55 89 144 - stopped being read as two phone numbers.

• Code samples are not contacts. Text inside <pre>, <code> and <samp> is left out of phone and email extraction, and a phone or an address published in the site's schema.org (JSON-LD) markup is read too.

• A date is not a phone, and neither is a parcel id. Publication dates (04.11.2026 19:48), map coordinates (50.4501 30.5234) and tracking numbers (1234 5678 9012 after the words "tracking number") are rejected before they can become a fake contact - every one of them was a live bug here, and every one now has a test.

• Dot-separated phones are understood. 202.555.0123 is a US number, not a decimal fraction; the guard that removed 3.14159 no longer eats it.

• A third-party mailbox still counts. If the only address on a site is a gmail.com inbox, it is returned in foreignEmails and the site counts as having contacts - the complaint we built this Actor against was "the email is on the site and the scraper misses it".

• Cloudflare and friends. A site that blocks automated access is reported as BLOCKED with the HTTP status in reason - and it is free.

• Same answer twice is cached. For cacheMinutes (default 24h) a repeated run is served from the cache in the account of whoever runs the Actor; those rows are marked fromCache: true and are not charged - you paid for that crawl the first time.

Run it your way

From the Apify UI, from the API, or from an MCP client. One input is enough:

{ "websites": ["apify.com", "stripe.com", "figma.com"] }

Words that are not guessed here: every email in the row exists on the site you sent us.