Website Contact Details Crawler
Pricing
from $0.005 / actor start
Crawl any list of websites and extract every email address, phone number (E.164) and social media profile, deduplicated into one clean record per domain. Fast HTTP-only crawler with pay-per-result pricing.
All notable changes to this Actor are documented here. The format follows Keep a Changelog and the project uses Semantic Versioning .
CheerioCrawler that crawls each website (same domain by default) and returns one deduplicated record per domain: emails, phones (E.164 + raw + country), social profiles, organisation name, address and a per-value sources map.mailto: links (entity- and percent-encoded), visible text, meta/title/aria-label attributes, Cloudflare data-cfemail / /cdn-cgi/l/email-protection#… decoding, and [at]/(dot)/name at domain dot com obfuscation.libphonenumber-js (full "max" metadata for accurate validation) from tel: links, JSON-LD and formatted numbers in text, using phoneCountryHint for national formats.schema.org Organization / LocalBusiness (and ~60 subtypes) JSON-LD parsing, including @graph, contactPoint, sameAs and lenient parsing of slightly malformed blocks.actor-start, page-crawled, domain-with-contacts (see .actor/pay_per_event.json).STATS (logged and persisted every 30 s) with categorised error counters, and FAILED_DOMAINS in the default key-value store.Actor.useState.apify/actor-node:20 base, but Node 20 reached end-of-life in April 2026 and gets no security fixes. Node 22 still meets "Node.js 20+".respectRobotsTxt: true, user-toggleable). The owner chose this for Store/ToS safety. Crawlee's built-in respectRobotsTxtFile is used; skipped pages are counted in STATS.pagesSkippedRobotsTxt.hasContacts: false) because their page-crawled events were already paid for. Only the domain-with-contacts value event depends on finding an email or phone. Websites where no page loaded are not pushed (nothing was charged for them); they are listed in FAILED_DOMAINS instead. This follows the "never emit an item you did not charge for" rule: every dataset row is backed by at least one charged page-crawled event.maxResults counts delivered websites, with or without contacts. Websites are fed into the queue in a window (at most maxResults − delivered and 25 at a time), so the run never crawls more websites than it can deliver.domain-with-contacts event for every website in progress. Users therefore never pay for pages of a website whose record could not be delivered. When the budget can't cover another website, it stops starting new ones and finishes the ones in flight. When it can't cover another page, the crawler stops. All charges go through a single mutex.calculateMaxEventChargeCountWithinLimit prices every event at $1. The Actor computes the same formula from the configured prices when not on the platform, and on the platform takes the minimum of both.prioritizePaths (whole path segment, so /en/contact-us/ matches) go to the front of the queue. Literal priority paths (the first 3) are only guessed when the homepage links to none of them, which avoids a burst of 404s. 404s are free.sameDomainDelaySecs: 1 means at most about one request per second per website, while many websites are crawled in parallel._cf_chl_opt, "Just a moment…", DataDome, PerimeterX, Incapsula, SiteGround). It never keys on the word "captcha" alone, because contact forms often embed reCAPTCHA. A redirect from a public page onto a login page is treated as "login required" and is not retried. HTTP 401 is handled the same way.+). This removes order numbers, SKUs and prices. Numbers in tel: links and JSON-LD are accepted unformatted.requestsFromUrl). Private, link-local and internal hostnames are rejected as an SSRF guard. If a start page fails over HTTPS with a network error, it is retried once over HTTP.npm audit reports 13 moderate advisories. They all trace to one transitive dependency, stream-json@1.9.1 via @crawlee/core (GHSA-528h-pc64-c93x, a DoS in the pick/ignore/filter/replace filters). Crawlee only uses StreamArray to deserialise its own persisted state, not the vulnerable filters, and never on attacker-controlled input. The fix requires a major version bump that Crawlee has not adopted yet. Re-check when upgrading Crawlee.