Company Contact Finder

Company Contact Finder

Layered, cheapest-first contact discovery at a specific company. Apollo People Search is the primary path (free search, no email reveal); a keyless LinkedIn-profile web search is the degrade when no Apollo key is present.

When to use

• "Who are the partners / CFOs / VP Finance at [company]?"
• "Find the heads of engineering at [company] and their LinkedIn."
• Downstream of qualification, when a company is ICP-fit and you need the buyer.

How to run

Primary — Apollo (when APOLLO_API_KEY is set)

python3 ${SKILL_DIR}/scripts/apollo_people.py \
  --company "Acme Corp" --domain acme.com \
  --titles "CFO,VP Finance,Head of Finance" \
  --num-results 10 \
  --output ${WORKSPACE}/contacts.json

--domain is optional; without it the script resolves the company name to a domain via Apollo Org Search. Free People Search returns name/title/LinkedIn/location; email is empty (reveal it later with apollo-lead-finder's enrich phase if needed).

Degrade — keyless (no Apollo key)

python3 ${SKILL_DIR}/scripts/serp_people.py \
  --company "Acme Corp" \
  --titles "CFO,VP Finance" \
  --num-results 10 \
  --output ${WORKSPACE}/contacts.json

Runs site:linkedin.com/in "<company>" "<title>" searches and returns candidate profile URLs + result title/snippet. You (the agent) resolve name from the snippet.

Select & dedup (you, the agent)

Read the JSON, dedup by normalized LinkedIn URL, drop off-target titles, trim to num_results, and return a contact table. If company_linkedin_url was provided to disambiguate a common name, prefer profiles whose company matches it.

Outputs

contacts.json — array of {name, title, company, company_domain, linkedin_url, location, email, source}. email is empty on both paths (use an enrichment skill to reveal); LinkedIn-only contacts are returned rather than dropped.

Credentials / env

• Required: none — the keyless serp path runs with no key.
• Optional: APOLLO_API_KEY (primary cheapest people search). DROPCONTACT_API_KEY and PHANTOMBUSTER_API_KEY (+ LinkedIn cookie) are deeper fallbacks the agent can invoke via their own skills when the two bundled paths under-deliver.

Notes & edge cases

• Layer cheapest-first: run Apollo when keyed; only fall to serp when it isn't, or to fill a shortfall. Stop once num_results is met.
• The keyless serp path uses a public search endpoint that rate-limits; in production this maps to the Robomotion serp Search node + proxy. Space queries; expect lower precision.
• Proxy + throttle any LinkedIn scraping; respect provider rate limits.
• Return LinkedIn-only contacts rather than nothing when enrichment is absent.