In this video, I show you how I built a self-healing automation system using n8n and Claude Code. Whenever one of my n8n workflows throws an error, it automatically triggers an error workflow that calls Claude Code. Claude then uses its n8n MCP server to audit the broken workflow, understand what went wrong, and fix it, all without any manual intervention. I just get a notification that the error was caught and resolved. The next time the workflow runs, it works perfectly because Claude already patched it. It's like having an AI engineer on call 24/7 to maintain your automations.

https://www.loom.com/share/a32d04f79ab94269ae65a0af337aad02 https://dashboard.n8n-mcp.com/ https://github.com/czlonkowski/n8n-mcp/tree/main This setup is designed for users who want to use Google’s Antigravity (Gemini) as their primary AI workspace without relying on Anthropic-specific tools. It solves the painstaking "Connection Closed: EOF" and "401 Unauthorized" errors that occur when a strict AI agent tries to maintain a real-time stream (SSE) through multiple proxy layers. Phase 1: Docker Layer (The Engine) The goal here is to stop n8n from "processing" the stream before it leaves your Pi. Below is a structured "Deep Dive" blueprint designed to be a definitive resource for self-hosting **n8n + MCP**. --- ## 🏗️ The "Direct Pipe" Architecture Guide This setup ensures a 100% stable, unbuffered stream between your Raspberry Pi and Google Antigravity. ### **Phase 1: The Docker Engine (.env & Compose)** The goal here is to stop n8n from "over-processing" the data before it leaves the Pi. * **Key `.env` Nuances**: * `N8N_SKIP_RESPONSE_COMPRESSION=true`: **Crucial.** Prevents n8n from gzipping the stream, which causes "Connection Closed: EOF" errors in strict AI agents. * `N8N_TRUST_PROXY=all`: Tells n8n to trust the headers (like `Authorization`) passed by Nginx and Cloudflare. * `N8N_CORS_ALLOWED_ORIGINS=https://n8n.yourdomain.com`: Prevents security rejections when the agent initiates the handshake. * **Docker Compose Secret**: * Ensure your `N8N_ENCRYPTION_KEY` is at least 32 characters. If you recreate your container without this being static, your MCP tokens will instantly become invalid. --- ### **Phase 2: The Nginx Location Block (The Pipe)** Standard Nginx settings "buffer" data, which is death for real-time MCP streams. You need a dedicated location block for the `/mcp-server/` path.

I’ve noticed that many people here are excited to get started, but at the same time feel unsure about where to focus first or whether they’re even heading in the right direction. I remember being in that exact spot myself. Sometimes it’s not about doing more, it’s about getting clarity on the next right step. What’s the one thing that’s currently holding you back the most? • Not knowing what to start with• Too many options and ideas • Lack of confidence • Information overload • Or something else? Drop it in the comments, chances are someone here has already faced it and found a way through. Let’s help each other move forward.

Just built a Lead Response & Qualification Automation Agent — this one literally behaves like a sales rep who never sleeps. Here's what happened when I deployed it for a property investment client: The Problem: Lead comes in at 9:52 PM. Team replies at 3:41 AM. CRM note: "Prospect signed with competitor." $14,000 deal. Gone. This was happening every single week. Here's how the whole system works: 1. Real-Time Lead Detection The agent monitors your lead sources 24/7: → Web forms → SMS inquiries → Email submissions → Chatbot conversations No manual checking. No missed notifications. 2. Configuration Layer You set once: → CRM API credentials → Qualification questions → Response templates → Escalation rules → Business hours vs after-hours behavior From there, it runs itself forever. 3. Instant Response Engine The moment a lead hits: → Agent replies in under 90 seconds → Personalized message (not generic) → Asks 2-3 qualifying questions → Logs everything in CRM Your prospects never wait. Ever. 4. Intelligent Qualification Based on responses, the agent determines: → Is this high-intent or tire-kicker? → Do they match our ICP criteria? → What's their timeline? → What's their budget range? Everything becomes structured data. 5. Smart Routing Then it splits into two paths: ✓ Path A — High-Intent Lead The agent: → Books them directly into calendar → Sends confirmation + prep materials → Notifies your closer immediately → Adds to high-priority pipeline ✓ Path B — Low-Intent or Unqualified The agent: → Sends nurture sequence → Tags for future follow-up → Keeps them warm (not cold) → Saves team from wasting time The Results After 30 Days: → 3 recovered deals that would've been lost → $42,000 in additional revenue → Zero team burnout → 100% response rate (even at 2 AM) Same team. Same ad spend. Just faster responses. Questions: → How effective can this be (ROI)? For this client: $42K revenue from 3 deals in 30 days. System cost: ~$2K to build + $200/month to run. ROI: 2,000%+ in first month.