n8n × AI Agent Integration— Building Local Workflow Automation with Claude and Gemini

Setting Up Your Self-Hosted Environment

Fastest Setup with Docker Compose

# docker-compose.yml
version: '3.8'
 
services:
  n8n:
    image: docker.n8nio/n8n:latest
    restart: always
    ports:
      - "5678:5678"
    environment:
      - N8N_BASIC_AUTH_ACTIVE=true
      - N8N_BASIC_AUTH_USER=admin
      - N8N_BASIC_AUTH_PASSWORD=your_secure_password
      - N8N_HOST=your-domain.com
      - N8N_PORT=5678
      - N8N_PROTOCOL=https
      - WEBHOOK_URL=https://your-domain.com/
      - GENERIC_TIMEZONE=Asia/Tokyo
    volumes:
      - n8n_data:/home/node/.n8n
      - /local-files:/files
 
volumes:
  n8n_data:

# Start
docker compose up -d
 
# Check logs
docker compose logs -f n8n

That's it — n8n is available at http://localhost:5678. For production, add Nginx as a reverse proxy and enable HTTPS with Let's Encrypt.

Production Deployment on a $6/Month VPS

My current setup: DigitalOcean $6/month Droplet (1vCPU / 1GB RAM) + Nginx + Let's Encrypt. n8n itself can run on 512MB RAM, but if you're processing large datasets through AI nodes, 1GB or more is recommended.

# Nginx config (/etc/nginx/sites-available/n8n)
server {
    server_name your-domain.com;
    
    location / {
        proxy_pass http://localhost:5678;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection 'upgrade';
        proxy_set_header Host $host;
        proxy_cache_bypass $http_upgrade;
        # Longer timeout for webhook processing
        proxy_read_timeout 300s;
    }
}

AI Agent Integration Patterns

n8n's AI nodes expanded significantly in late 2024. You can now integrate Claude, Gemini, and OpenAI as workflow nodes, with full AI agent orchestration built into n8n itself.

Claude Integration Setup

Register your Anthropic API key in n8n Credentials, then add an "Anthropic Chat Model" node to your workflow. Here's an example of generating dynamic prompts with n8n's Code node:

// Generate dynamic prompt in n8n Code node
const emailBody = $json.email_body;
const sender = $json.sender_name;
 
return {
  prompt: `Classify the following email.
  
Sender: ${sender}
Body: ${emailBody}
 
Return category (support/sales/other) and priority (high/medium/low) as JSON.
Format: {"category": "...", "priority": "...", "summary": "..."}`
};

Real Workflow 1: Automated Email Classification

This is a workflow I actually run in production:

Gmail Trigger (new email)
  │
  ├─ Filter: Remove spam and promotions
  │
  ├─ Claude Node: Analyze and classify email content
  │   Prompt: Return category, priority, summary as JSON
  │
  ├─ Switch Node: Branch by category
  │   ├─ Support → Create ticket in Notion database
  │   ├─ Sales → Post to Slack #sales channel
  │   └─ Other → Apply Gmail label only
  │
  └─ Completion notification

The key implementation detail: specify the return format as strict JSON. Vague format instructions will cause parsing failures in downstream nodes.

Real Workflow 2: Weekly Report Auto-Generation

Scheduler (every Monday at 9:00)
  │
  ├─ Google Analytics Node: Fetch last week's traffic data
  ├─ Google Search Console: Fetch search performance data
  ├─ Notion Node: Fetch completed task count for last week
  │
  ├─ Code Node: Format data and compose prompt
  │
  ├─ Gemini Node: Generate report body
  │   (Gemini 2.5 Pro handles long-form generation well)
  │
  └─ Gmail Node: Send report to self

Real Workflow 3: SNS Post Scheduling

Notion Trigger (new content item added)
  │
  ├─ Claude Node: Optimize copy for each platform
  │   ├─ X (Twitter): Under 280 characters
  │   ├─ LinkedIn: Professional tone
  │   └─ Instagram: Hashtag optimization
  │
  ├─ Approval step (Review in Slack → Button approval)
  │
  └─ Buffer / Hootsuite Node: Schedule posts to each platform

Using n8n's AI Agent Mode

Recent versions of n8n include an "AI Agent" node that goes beyond simple LLM calls to enable autonomous agents within your workflows:

AI Agent Node
  ├─ System Prompt: "You are a data analysis agent. Use tools to..."
  ├─ Tools: Calculator, HTTP Request, Code Executor
  └─ Model: Claude 3.5 Sonnet

The agent decides on its own: "I need to calculate this, so I'll use the Calculator" or "I need more data, so I'll fetch via HTTP." I use this for content research collection workflows.

3 Security Lessons I Learned the Hard Way

1. Randomize your webhook URL paths Leaving the default /webhook/ path exposed will get it hammered by scanners. Use random strings like /webhook/a7f3k9m2b6/.

2. Manage API keys through Credentials or environment variables Never hardcode API keys directly inside workflow nodes. Use n8n's Credentials feature or .env files.

3. Set up regular backups The n8n_data volume contains all your workflows and credentials. Run a separate n8n workflow to back it up to S3 or Backblaze B2 every day.

A Note from an Indie Developer

Where to Go Next

Once your n8n foundation is solid, try building custom nodes. Any service not covered by existing integrations can be wrapped as an n8n node in Node.js. Also worth evaluating: n8n Cloud (paid, but zero maintenance) vs. self-hosted — weigh the cost against the ops burden based on your situation.

Automation isn't "set it and forget it" — it's something you grow over time. Start small, verify behavior, and gradually add complexity. That's the approach that actually sticks.

Workflow examples in this guide were tested on n8n v1.x. UI and node names may differ across versions.