Feature

Let an AI agent call the deliverability engine directly.

@cognilead/mcp-server publishes CogniLead as a set of callable tools over the Model Context Protocol — four work with no API key so an agent can learn the surface immediately, twelve more unlock the real pipeline once you hand it a key. This page is a quick overview; see /mcp for the full tool, Resource, and Prompt reference.

Start free →See per-send pricing →

mcp-server

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

Start free →Read the full spec →

The problem

What makes this hard today.

Agents cannot click a dashboard

An AI agent drafting and sending outbound needs a callable tool interface, not a UI built for a human operator.

Handing an agent unrestricted send access is risky

Letting an agent send real cold email without guardrails risks reputation damage or compliance violations at machine speed.

Most deliverability tools have no MCP surface at all

Plenty of cold-email tools are UI-only or API-only, with nothing an MCP-compatible IDE or agent framework can call natively.

How CogniLead helps

The mechanism, step by step.

1
Four keyless tools to start
draft_pitch, score_signal, check_deliverability, and health all work with no API key — an agent can learn the surface before you commit a key.
2
Twelve gated tools for the real pipeline
ingest_lead, suppress, generate_seo_content, and read-only inspection tools (leads, sends, suppressions, billing, subscription, webhooks, keys) unlock with a key.
3
Guardrails hold regardless of caller
Every send through the agent still passes suppression and reputation checks — the agent cannot opt out of the same guardrails the REST API enforces.

FAQ

Questions about MCP server.

How many tools does the MCP server expose?+

16 total — 4 keyless, 12 gated — plus 4 Resources and 3 Prompts. See the full table at /mcp.

Can an agent send real outbound through MCP?+

Yes, once given an API key — ingest_lead and the send pipeline are gated tools that run through the same suppression and circuit-breaker checks as the REST API.

How do I install it in my IDE?+

npx -y @cognilead/mcp-server, configured in your IDE’s MCP config file — see the Cursor, Claude Desktop, and Windsurf snippets on /mcp.