- Remove code-focused tool imports (analyze, codereview, debug, etc.) - Keep only marketing tools: chat, contentvariant, listmodels, version - Add TODO comments for thinkdeep and planner (will copy later) - Add git workflow commands to CLAUDE.md with Forgejo remote details
4.5 KiB
4.5 KiB
Zen-Marketing MCP Server - Setup Complete
✅ What's Been Done
1. Core Architecture ✓
- Copied essential files from zen-mcp-server
- Adapted for marketing-focused workflows
- Preserved conversation continuity, multi-model support, and file handling
2. Configuration ✓
config.py- Marketing-specific configuration with temperature profiles.env.example- Template for API keys and settings- Model configuration for:
google/gemini-2.5-pro-latest(analytical/strategic)google/gemini-2.5-flash-preview-09-2025(fast generation)minimax/minimax-m2(creative content)
3. Tools Implemented ✓
- contentvariant - Generate 5-25 content variations for A/B testing
- chat - Marketing strategy and brainstorming
- listmodels - Show available AI models
- version - Server version and info
4. Infrastructure ✓
server.py- Main MCP server with tool registration- Provider system (Gemini + OpenRouter)
- Logging system (file + console)
run-server.sh- Setup and launch script
5. Git Repository ✓
- Initialized with clean commit history
- Remote configured:
ssh://git@git.tealmaker.com:2222/ben/zen-marketing.git - Ready to push when network accessible
📋 Next Steps
1. Push to Forgejo (Manual)
cd ~/mcp/zen-marketing
git push -u origin main
2. Configure API Keys
cd ~/mcp/zen-marketing
cp .env.example .env
# Edit .env with your API keys:
# - OPENROUTER_API_KEY (for minimax and others)
# - GEMINI_API_KEY (optional, for direct Gemini access)
3. Test Locally
cd ~/mcp/zen-marketing
./run-server.sh
4. Configure Claude Desktop (MacBook)
Add to ~/.claude.json:
{
"mcpServers": {
"zen-marketing": {
"command": "/path/to/zen-marketing/.venv/bin/python",
"args": ["/path/to/zen-marketing/server.py"],
"env": {
"OPENROUTER_API_KEY": "your-key",
"DEFAULT_MODEL": "google/gemini-2.5-pro-latest",
"FAST_MODEL": "google/gemini-2.5-flash-preview-09-2025",
"CREATIVE_MODEL": "minimax/minimax-m2",
"ENABLE_WEB_SEARCH": "true",
"LOG_LEVEL": "INFO"
}
}
}
}
Then restart Claude Desktop.
5. Test the contentvariant Tool
In Claude Desktop:
Use zen-marketing contentvariant to generate 10 subject line variations
for an HVAC newsletter about PCB diagnostics. Test angles: technical
curiosity, contrarian, knowledge gap, urgency.
🎯 Priority Tools to Implement Next
Based on PLAN.md priority:
- subjectlines - Specialized email subject line generator (High Priority)
- platformadapt - Cross-platform content adaptation (High Priority)
- styleguide - Writing style enforcement (High Priority)
- factcheck - Technical verification (High Priority)
- seooptimize - WordPress SEO workflow (Medium Priority)
- guestedit - Guest content editing (Medium Priority)
📁 File Structure
zen-marketing/
├── server.py # Main MCP server
├── config.py # Configuration
├── requirements.txt # Dependencies
├── .env.example # Environment template
├── run-server.sh # Setup script
├── tools/
│ ├── chat.py # Marketing chat
│ ├── contentvariant.py # Variation generator ✨ NEW
│ ├── listmodels.py # Model listing
│ └── version.py # Version info
├── systemprompts/
│ ├── chat_prompt.py
│ └── contentvariant_prompt.py ✨ NEW
├── providers/ # AI provider integrations
├── utils/ # Utilities
└── logs/ # Log files
🔧 Model Configuration
Default Model Strategy:
- Gemini 2.5 Pro - Analytical work (styleguide, seooptimize, guestedit)
- Gemini Flash - Fast generation (subjectlines, quick variations)
- Minimax M2 - Creative content (contentvariant, platformadapt)
All models accessible via OpenRouter with single API key.
📝 Testing Checklist
- API keys configured in .env
- Local test:
./run-server.shruns without errors - Claude Desktop can see zen-marketing server
- contentvariant tool generates 10+ variations
- Variations respect platform character limits
- Chat tool responds with marketing context
🚀 Ready for Phase 2
The foundation is complete. Ready to implement the high-priority simple tools:
- subjectlines
- platformadapt
- styleguide
- factcheck
All following the established patterns in contentvariant.py.