Lead Research Factory

{
  "mcpServers": {
    "lead-research-factory": {
      "command": "/bin/bash",
      "args": [
        "/Users/bizrate/Documents/ChatGPT/Research_factory/scripts/start-mcp.sh"
      ],
      "env": {
        "API_BASE_URL": "http://localhost:4000"
      }
    }
  }
}

Use the lead-research-factory MCP server as the control plane for lead research.

Operating model:
- Codex is the research director.
- The dashboard is for monitoring, settings, proxies, progress, and CSV export.
- Qwen and workers do bulk research.
- Codex should use MCP tools instead of manually operating browsers.

Normal workflow:
1. Check get_runtime_settings, system_health, and system_stats before creating a campaign.
2. Update Redis, Qwen, storage, browser, capacity, and provider runtime settings with update_runtime_settings when the user asks.
3. Create campaigns from the user's ICP prompt with create_campaign.
4. Use strict evidence. Unknown is better than guessed.
5. Inspect reusable providers with list_source_recipes, list_enrichment_providers, and list_email_verification_providers before creating new ones.
6. When a source needs custom browser handling, start Camoufox with start_debug_browser and inspect it with debug_browser_snapshot/open/click/type/extract/screenshot.
7. Convert the debug trail into a source recipe draft with debug_browser_recipe_draft, then save it with create_source_recipe.
8. When an HTTP API can enrich company/contact data, save it with create_enrichment_provider. Store reusable provider credentials in provider templates or server secrets.
9. When an HTTP API can verify discovered emails, save it with create_email_verification_provider. Map response fields so workers can mark valid/risky/invalid emails.
10. Reuse active providers before creating new ones. Use trial providers for one campaign, and global active providers when reusable.
11. Inspect leads, evidence, provider runs, and weak fields before presenting recommendations.
12. Pause, resume, cancel, audit, rerun analysis, and export through MCP when those tools are available.

Currently wired MCP tools:
- get_runtime_settings
- update_runtime_settings
- create_campaign
- create_source_recipe
- list_source_recipes
- get_source_recipe
- activate_source_recipe
- disable_source_recipe
- create_enrichment_provider
- list_enrichment_providers
- get_enrichment_provider
- activate_enrichment_provider
- disable_enrichment_provider
- create_email_verification_provider
- list_email_verification_providers
- get_email_verification_provider
- activate_email_verification_provider
- disable_email_verification_provider
- start_debug_browser
- list_debug_browsers
- debug_browser_snapshot
- debug_browser_open
- debug_browser_click
- debug_browser_type
- debug_browser_extract
- debug_browser_screenshot
- debug_browser_recipe_draft
- close_debug_browser
- system_stats
- parse_proxies
- upload_proxies
- list_campaigns
- get_campaign
- pause_campaign
- resume_campaign
- cancel_campaign
- get_leads
- get_lead_evidence
- export_campaign_csv
- system_health
- list_proxies

App endpoints:
- Dashboard: http://localhost:3000
- API base: http://localhost:4000
- MCP stdio launcher: /Users/bizrate/Documents/ChatGPT/Research_factory/scripts/start-mcp.sh