Lead Generation (Exa)

Turn an ICP into a prospect list: companies → decision-makers → enrichment. This is a multi-pass workflow built on exa-company-research and exa-people-research. Calls the Exa REST API through local scripts; no MCP server needed, only an EXA_API_KEY.

Setup (once)

bash
export EXA_API_KEY=your-key            # or EXA_API_KEY=... in a .env at the repo root

Get a key at https://dashboard.exa.ai/api-keys. Shared details: exa-native-base.

The pipeline

Pass 1 — Source companies (this skill's script):

bash
python scripts/lead_generation.py "Series A B2B SaaS companies in fintech" -n 20
python scripts/lead_generation.py "category:company developer tools startups hiring backend 2026" -n 20

Run 2-3 angles (category, stage, recent-launch) and dedupe to a clean company list.

Pass 2 — Find decision-makers at each company (use exa-people-research, batch 3-5 companies per subagent):

bash
python ../exa-people-research/scripts/people_research.py "VP Engineering at <Company>" -n 5
python ../exa-people-research/scripts/people_research.py "Head of Sales at <Company>" -n 5

Pass 3 — Enrich (funding, hiring, news signals → timing):

bash
python ../exa-company-research/scripts/company_research.py "<Company> funding round hiring news" -c news -n 5

Token isolation (essential here)

This is a fan-out task. One subagent per batch of seeds (3-5 companies each), each running Passes 2-3 and returning a compact table. Then compile, dedupe by company + person, and rank. Never run the whole enrichment loop in your main context. See exa-native-base for the orchestration pattern.

Qualifying & enriching rows

• Define what makes a valid lead before searching (stage, geo, headcount, tech stack, buying signal).
• Capture a consistent schema per row: company · website · why-it-fits · contact name · role · profile URL · signal (hiring/funding/launch) · source.
• Apply hard filters first (stage, geo), then soft filters (genuine fit). See references/filtering.md in exa-native-base.

Finding hidden connections

For "who are X's customers / partners", direct queries return articles, not connections. Use indirect signals:

bash
python ../exa-company-research/scripts/company_research.py "<Company> case study customer success story" -n 5
python ../exa-native-base/scripts/exa.py contents https://company.com/customers https://company.com/case-studies --text

After you get results

Deliver a deduped table (company · fit · contact · role · profile · signal · source), lead with the highest-signal accounts, and note coverage gaps. Treat results as similarity, not validation — verify before outreach.