exa-lead-gen

name	exa-lead-gen
description	Generate enriched lead lists using Exa deep search. Finds companies matching an ICP, enriches with signals/news/scores, and outputs CSV. Use when generating leads, building prospect lists, finding companies to sell to, doing outbound research, or ICP-based company discovery.
user-invocable	true

Lead Generation (Exa Deep Search)

Tool

Call the native MCP tool exa-deep__deep_search_exa directly. Parameters go in a JSON object.

Architecture

Main agent orchestrates; subagents run the searches. This keeps context lean.

Main Agent (orchestrator)
├── Step 1: ICP research (1 deep call)
├── Step 2: Generate micro-verticals (LLM reasoning)
├── Step 3: Design outputSchema
├── Step 4: Batch subagents (5 micro-verticals each, parallel)
│     Each subagent: runs 5 deep calls → writes JSON → reports count
├── Step 5: Python CSV compiler (reads JSON, dedupes, sorts)
└── Step 6: Summary

Required Params on Every deep_search_exa Call

structuredOutput=true
numResults=50
highlightMaxCharacters=1
type=deep

numResults and systemPrompt must align — set numResults=50 and ask for "exactly 50 companies" in systemPrompt.

outputSchema Constraints

• Max 10 properties total across all nesting levels
• Array items: flat objects with primitive fields only (string, integer, boolean, array of strings)
• Every string field must have a word limit in its description
• Root must be "type": "object"

Call Examples

Tool: exa-deep__deep_search_exa

# Step 1: ICP Research
{ objective: "About {company_name}, {company_name} customers",
  systemPrompt: "Research the company and return ICP, sub-verticals, and useful enrichments",
  structuredOutput: true, numResults: 10, highlightMaxCharacters: 1, type: "deep" }

# Step 4: Lead gen call (inside subagent)
{ objective: "B2B sales intelligence platforms using web scraping and contact data enrichment",
  systemPrompt: "List exactly 50 companies. Score each 1-10 on ICP fit. Return structured enriched data.",
  structuredOutput: true, numResults: 50, highlightMaxCharacters: 1, type: "deep" }

Micro-Vertical Budget

Target ceil(requested_leads / 35) micro-verticals (overshoot for dedupe losses).

Step 1: ICP Research Schema

{
  "type": "object",
  "properties": {
    "company_description": {"type": "string", "description": "What they do in 2 sentences"},
    "product_description": {"type": "string", "description": "What they sell and to whom, 2 sentences"},
    "existing_customers": {"type": "array", "items": {"type": "string"}},
    "icp_description": {"type": "string", "description": "ICP in 12 words or less"},
    "sub_verticals": {"type": "array", "items": {"type": "string"}},
    "demographic_signals": {"type": "array", "items": {"type": "string"}},
    "useful_enrichments": {"type": "array", "items": {"type": "string"}}
  }
}

Standard Lead Schema (customize per use case)

{
  "type": "object",
  "properties": {
    "companies": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "company_name": {"type": "string", "description": "in 5 words or less"},
          "website": {"type": "string", "description": "homepage URL"},
          "product_description": {"type": "string", "description": "in 12 words or less"},
          "icp_fit_score": {"type": "integer"},
          "icp_fit_reasoning": {"type": "string", "description": "one-liner in 20 words or less"},
          "industry_vertical": {"type": "string", "description": "in 3 words or less"},
          "funding_stage": {"type": "string", "description": "Bootstrap/Seed/Series A/B/C+/Public/Unknown"},
          "headquarters_location": {"type": "string", "description": "City, Country in 4 words or less"},
          "recent_signals": {"type": "array", "items": {"type": "string", "description": "under 12 words each"}}
        }
      }
    }
  }
}

Output Format

After CSV is generated, report:

• Total leads, duplicates removed
• ICP score distribution (8-10 / 5-7 / 1-4)
• Number of Exa calls made
• Output filename

Performance

• Each deep_search_exa call: 4-12s
• Yield: ~35-48 companies per call (avg ~42)
• For 500+ leads: confirm with user before starting
• Launch batch subagents in waves of ~6 to respect QPS limits