FeedMob AdPilot MCP Server

AI-powered advertising campaign planning and creative generation system built with FastMCP, Claude Agent SDK, and mcp-ui.

Features

🎯 Parse Advertising Requirements

Extract structured campaign parameters from natural language descriptions.

🔍 Conduct Ad Research

Generate comprehensive campaign reports with market insights, audience analysis, and platform strategies.

✍️ Generate Ad Copy

Create two distinct ad copy variations optimized for your target platform and audience.

🖼️ Generate Ad Images

Generate image variations for campaigns with interactive selection.

🎨 Generate Mixed Media

Combine selected images with ad copy for complete creative assets.

📊 Campaign Management

Store and retrieve campaign data with PostgreSQL persistence.

Quick Start

Prerequisites

• Node.js 20+
• Docker (for PostgreSQL)
• AWS Account with Bedrock access (for Claude Agent SDK)

Installation

npm install

Configuration

cp .env.example .env

Configure your .env file with:

• AWS credentials for Bedrock
• PostgreSQL connection (or use Docker)
• Optional: Tavily API key for enhanced research

Start Database

docker-compose up -d postgres

Run Development Server

npm run dev

Server starts at http://localhost:8080/mcp

Test with MCP Inspector

npm run mcp:inspect

Available Tools

Tool	Description
parseAdRequirements	Parse natural language campaign requirements
conductAdResearch	Generate comprehensive campaign research reports
generateAdCopy	Create ad copy variations
generateAdImages	Generate image variations
generateMixedMediaCreative	Combine images and copy
getCampaign	Retrieve stored campaign data

Project Structure

├── src/
│   ├── index.ts              # FastMCP server entry
│   ├── tools/                # MCP tool definitions
│   ├── services/             # Agent services (Claude SDK)
│   ├── utils/                # UI factories (mcp-ui)
│   ├── schemas/              # Zod validation schemas
│   └── plugins/              # Claude Agent skills
├── client-ui/                # Next.js chat interface
├── tests/                    # Unit and property tests
├── docs/                     # Documentation
└── .kiro/specs/              # Feature specifications

Endpoints

• MCP: http://localhost:8080/mcp
• SSE: http://localhost:8080/sse
• Health: http://localhost:8080/health

Development

npm run dev          # Start with hot reload
npm run build        # Build for production
npm start            # Run production build
npm test             # Run tests
npm run typecheck    # Type checking
npm run mcp:inspect  # Visual debugging

Docker Deployment

See DEPLOYMENT.md for Coolify/Docker deployment instructions.

Pre-built images available at:

• ghcr.io/feedmob/feedmob_adpilot_mcp:latest
• ghcr.io/feedmob/feedmob_adpilot_mcp-ui:latest

Technology Stack

• FastMCP - MCP server framework
• @mcp-ui/server - Interactive UI components
• @anthropic-ai/claude-agent-sdk - Claude Agent integration
• Zod - Runtime type validation
• PostgreSQL - Campaign data persistence
• Vitest + fast-check - Testing

Client UI

The client-ui/ directory contains a Next.js chat interface that connects to this MCP server. See client-ui/README.md for details.

License

MIT