Website Contact Extraction

Website Contact Extraction

Take a list of website URLs and return a spreadsheet with each site's phone number(s) and email(s), pulled from their contact/about pages (or homepage, if that's where the info lives).

Workflow

1. Get the input list

• Uploaded Excel/CSV: read it (the xlsx skill's approach applies for reading). Find the column containing URLs — if the file has other columns (company name, etc.), keep them and add the new contact columns alongside.
• Pasted list: use as given, one URL per line/entry.

Normalize each URL before fetching (add https:// if missing, strip trailing whitespace). Keep the original URL/domain as given by the user as the row identifier even if you're redirected elsewhere while fetching.

2. Find contact info for each site

For each site, in order (stop as soon as you've found solid contact info):

1. Fetch the homepage directly (web_fetch — this is fine since the URL came from the user's own list). Many sites list a general email/phone right in the header or footer.
2. Look for mailto: and tel: links in the fetched page content — these are the most reliable signal since the site itself marked them as contact methods, not just text that happens to look like an email/phone.
3. If the homepage doesn't have enough, find the contact/about page. Prefer discovering its URL via web_search (e.g., "<domain> contact us") so the returned URL is fetchable, rather than guessing a path like /contact-us directly — web_fetch can only open URLs that already appeared in search results or fetched content, so a guessed path may not be fetchable. Then fetch that discovered contact/about page.
4. If nothing is found after the above, mark this site as not found (see step 4) rather than continuing to guess further paths.

See references/finding-contact-info.md for extraction patterns (email/phone regex considerations, filtering out junk matches like template placeholder emails or tracking-pixel addresses) and for known limitations (JS-rendered sites, obfuscated emails).

3. Extract and dedupe

Pull every genuine email and phone number found across the pages checked for that site. Dedupe (the same email/phone often appears in both header and footer). Combine multiple values into one comma-separated string per field, per the user's format preference — do not fabricate or guess at a value that wasn't actually present on the page.

4. Handle sites where nothing is found

Keep the row in the output file, with the Emails/Phone Numbers fields left empty for that site — don't drop the row and don't fill it with a placeholder like "Not Found." Still track the count and reasons (no contact info on any page checked / page didn't load / site blocked fetching) and report that in the chat summary, so the user knows how many came back empty and roughly why.

5. Build and deliver the output

Build the result as a CSV or Excel file (match whatever the user's input was, or ask if input was pasted text) with the xlsx skill's approach. Columns: the original identifying info (URL/domain, plus any columns that were already in an uploaded file), Emails, Phone Numbers. Save to /mnt/user-data/outputs/ and present it.

Notes

• This only works on publicly displayed contact information (a site's own contact/about page) — it's the same kind of extraction a person would do manually, just faster; it does not access anything behind a login or paywall.
• For large lists, process steadily and give the user a sense of progress if it's a long list rather than going silent — a brief "working through N sites" is fine.
• If a site is a single-page app or heavily JS-rendered, web_fetch may only see a shell of the page with no visible contact info even though a browser would show it — note this as a likely cause when a well-known company's site turns up nothing.
• Never invent or guess an email/phone number (e.g., pattern-guessing info@domain.com because that's common) unless it was actually found on a fetched page.