Lead Enrichment

Lead finding + enrichment with webclaw

Start from whatever the user has — a CSV, a listing URL, or just a description of who they sell to — and end with an enriched lead sheet: what each company does, who it sells to, a published contact email, socials, pricing model, and the tech it mentions. Everything comes from each company's own public pages — no licensed contact database, no per-seat platform.

What this skill does (and what it doesn't)

This skill is the discovery + firmographics half of lead gen. It finds companies (from listings or an ICP) and reads generic facts off each company's own site via /v1/extract — pitch, a published contact email, socials, pricing model, target customer, tech signals, or any custom field you define. That flexible schema is its whole point.

It does not recover people. For founders/leadership with their LinkedIn and X, use webclaw's dedicated tools (see the webclaw skill):

• lead — one company URL → a nested lead object with founders/leadership (name, role, linkedin, x) plus firmographics, from open-web search + verification. Flat 100 credits per successful lead.
• lead_batch — up to 25 URLs, blocks until done, returns per-URL results. 100 credits per successful lead.

Pick by what you need per company:

A common flow: run find.py to build the list, then send it to lead_batch for contacts, or to enrich.py for custom firmographics — or both.

Prerequisites

The webclaw skill / MCP server (npx create-webclaw). For bulk runs, a WEBCLAW_API_KEY (free tier at webclaw.io works).

Pick the entry point by what the user has

Finding leads (scripts/find.py)

# from a listing page you already know (repeatable --url)
WEBCLAW_API_KEY=wc_... python3 scripts/find.py --url https://www.ycombinator.com/companies/industry/developer-tools

# from a plain-language ICP — searches the web, harvests the listicles it finds
WEBCLAW_API_KEY=wc_... python3 scripts/find.py --query "AI agent startups that read the web"

Writes leads.csv (name, website, one_liner, source), deduped across sources. "Best X tools" articles and directories are pre-curated lead lists — this harvests them. Note: directory pages (YC, PH) often yield profile URLs rather than company homepages; enrichment still works on those, but if the user wants homepage-level fields, extract the real website from each profile first.

Example: revenue-qualified leads from TrustMRR

trustmrr.com is a public database of 15,000+ startups with payment-provider-verified revenue, and it is deliberately AI-friendly (llms.txt, per-startup .md pages, an AI API). Its category pages make excellent find sources, and profile pages enrich into revenue-qualified leads:

python3 scripts/find.py --url https://trustmrr.com/category/ai --out leads.csv
python3 scripts/enrich.py leads.csv --schema revenue.json

with revenue.json asking for monthly_revenue, founder_name, founder_x_handle, and actual_website. Verified result in testing: MRR figures, founder names, X handles, and the startup's real site — a lead sheet where every row provably has revenue. If the user wants homepage-level fields too, run a second enrich pass over the actual_website column.

Two paths

Small lists (up to ~15 companies): drive the MCP tools directly

For each row, call the webclaw extract tool with the company URL and this schema, then assemble the results into a table or CSV yourself:

{
  "type": "object",
  "properties": {
    "company_name":    {"type": "string"},
    "one_line_pitch":  {"type": "string"},
    "contact_email":   {"type": "string", "description": "a published business contact email, empty string if none is public"},
    "social_links":    {"type": "array", "items": {"type": "string"}},
    "pricing_model":   {"type": "string", "description": "free / freemium / paid / enterprise / unknown"},
    "target_customer": {"type": "string"},
    "tech_signals":    {"type": "array", "items": {"type": "string"}, "description": "frameworks, APIs, and integrations the site mentions"}
  }
}

One call per company, a few seconds each. Runs locally where possible; bot-protected or JS-heavy sites escalate to the hosted engine when WEBCLAW_API_KEY is set.

Bulk lists (15+): run the bundled script

WEBCLAW_API_KEY=wc_... python3 scripts/enrich.py leads.csv

• Auto-detects the URL column (website, domain, url, ...), enriches concurrently, and writes leads-enriched.csv with the original columns plus the extracted fields and an enrich_error column for any row that failed.
• --limit 5 for a cheap test run first. --concurrency, --out as needed.
• Python stdlib only — no installs.

Custom fields

The fields are just a JSON schema, so change them to fit the play. Write the schema to a file and pass --schema:

python3 scripts/enrich.py leads.csv --schema hiring_signals.json

Examples that work well: is_hiring, has_api, uses_competitor_x, recent_funding_mentioned, languages_supported, has_free_tier. Anything a human could answer by reading the company's site, the extractor can answer too.

Workflow for the agent

1. Ask for (or locate) the input CSV and confirm which column holds the website.
2. If the user wants non-default fields, build the schema with them before running.
3. Small list → per-row extract calls; bulk → scripts/enrich.py with --limit 5 first, show the sample, then run the full list.
4. Summarize results: how many rows enriched, how many had a published email, notable patterns (pricing models, common tech).
5. Suggest next steps: sort by fit, or feed the output into their CRM.

Notes

• Respect privacy law. This extracts only what companies publish on their own sites, and it's the user's responsibility to use contact data lawfully (GDPR/CAN-SPAM). Enrich business sites, not personal pages.
• Expect some empty contact_email fields — many companies publish none. That's signal too. contact_email here is a published mailbox off the site (e.g. hello@company.com), not a person — for founders and their profiles, use lead / lead_batch.
• Cost: each company is one /v1/extract call (25 credits). Founder/contact recovery via the lead tool is separate and priced higher (a flat 100 credits per lead), since it does the open-web people search on top. No per-seat pricing either way.