MCP Server Overview

CampaignStack exposes its entire platform through the Model Context Protocol (MCP). The MCP server provides 55+ tools across 9 domains, giving AI agents full programmatic access to campaigns, leads, workflows, content, and more.

Architecture

AI Agent / MCP Client
        │
        ▼
  MCP Server (port 3002)
    ├── Auth: API key validation + scope check
    ├── Validation: Zod .strict() input schemas
    └── Backend: Convex (serverless)
              └── Runner (LinkedIn automation)

The server supports two transports:

• stdio — local development, no authentication
• HTTP — production, requires API key

Tool domains

Domain	Tools	Description
Campaigns	4	Create, list, get, update campaigns
Leads	3	Search, get, import leads
Lead Lists	4	Manage lead collections
Workflows	4	Trigger and manage automation sequences
ICPs	3	Ideal Customer Profile scoring
Content	14	LinkedIn post lifecycle (draft to published)
Craft	4	AI-powered message and note writing
Competitors	6	Company/profile monitoring
Workspace	7	Generation, scraping, diagnostics
Other	6	Enrichment, search, workflow nodes

Design principles

• JSON-only responses — all tool output is structured JSON, never markdown
• Cross-referenced errors — error messages suggest which tool to call next
• Strict validation — Zod .strict() rejects unknown input fields
• MCP annotations — every tool declares readOnlyHint, destructiveHint, idempotentHint
• Actionable descriptions — tool descriptions reference related tools to help agents chain operations

Endpoints

Method	Path	Auth	Purpose
GET	/health	None	Server status, version, tool list
POST	/mcp	Bearer	MCP JSON-RPC messages
GET	/mcp	Bearer	SSE stream for sessions
DELETE	/mcp	Bearer	Close session

Next steps

• Authentication — API keys and scopes
• Transports — stdio vs HTTP
• Tool Reference — Full tool documentation