exa-lead-genlisted

# Lead Generation (Exa Deep Search) ## Tool Use `exec` to call: `mcporter call exa-deep.deep_search_exa <params>` ## 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 Syntax ```bash # Step 1: ICP Research mcporter call exa-deep.deep_search_exa \ 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) mcpor