Cold email via MCP: let your AI agent send outbound

AI agents are getting good at the top of the outbound funnel: research a company, find the angle, draft the pitch. The gap is the last step — actually sending, without burning a domain or violating compliance. The Model Context Protocol (MCP) closes it by giving the agent real tools.

What the CogniLead MCP server gives an agent

• Free, no-key tools: draft_pitch, score_signal, check_deliverability (live SPF/DMARC grading), and health.
• Gated tools: ingest_lead and suppress drive the real pipeline today; run_campaign is registered on the surface but not live yet (it returns a structured "unavailable" error until campaign endpoints ship).
• Resources: live sends, suppression list, and warmed-pool state the agent can read.
• Prompts: cold-outbound-sequence, deliverability-triage, warmup-plan — playbooks that keep output compliant.

Install in your IDE

{
  "mcpServers": {
    "cognilead": {
      "command": "npx",
      "args": ["-y", "@cognilead/mcp-server"],
      "env": { "COGNILEAD_API_KEY": "pk_live_..." }
    }
  }
}

Omit the key and the three free tools still work — an agent can learn the surface before you commit. Add a key and it can run real campaigns through the warmed pool.

See the full MCP surface and setup on the MCP page.