Generate n8n workflows from chat using MCP tools, Claude and Postgres

Describe any automation in plain English. This AI agent builds,
validates, and deploys a working n8n workflow into your instance
in seconds — no node dragging, no JSON editing, no guesswork.

📺 Full setup guide: [https://youtu.be/HMXBInLaUB8?si=Q21gt48omSyzsTGl]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🎯 WHO IS THIS FOR?
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

→ n8n builders who want to generate workflows by chatting
instead of configuring every node manually
→ Agencies building automations for clients at speed
→ Developers who want an AI co-pilot with real n8n knowledge
→ Non-technical teams who need to create workflows by
describing them in plain language

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
⚡ WHAT MAKES THIS DIFFERENT
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Most AI workflow generators hallucinate node names, guess
parameters, and produce broken JSON. This one doesn't.

It uses the official n8n MCP Client Tool node — natively
built into n8n, no community packages required — to connect
to a self-hosted n8n-mcp server (czlonkowski/n8n-mcp, 16k+
GitHub stars). That server holds the complete, structured
documentation of 1,396 n8n nodes in a SQLite database with
12ms response time and zero rate limits.

The agent never guesses. It looks everything up first.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✅ WHAT IT DOES
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

→ Accepts plain English prompts via n8n's built-in chat UI
→ Looks up exact node type strings — never assumes from
training data
→ Validates every node config before building anything
→ Runs 3 separate post-build checks: structure, connections,
and expression syntax
→ Deploys directly to your n8n instance via the REST API
→ Returns only the workflow URL + credential checklist
→ Remembers conversation history for follow-up edits
→ Modifies existing workflows with diff updates, not rebuilds

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
⚙️ HOW IT WORKS — STEP BY STEP
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

1. Chat Trigger receives your message and assigns a session ID

2. Postgres Chat Memory stores conversation history so
   follow-ups like "change the trigger to a webhook" work
   naturally without losing context

3. AI Agent (Claude Sonnet 4.5) reads the system prompt
   and begins calling MCP tools in strict order:

   — tools_documentation() to load current best practices
   — search_nodes() for exact node type strings
   — get_node_essentials() for the 10-20 params that matter
   — search_templates() to find proven real-world patterns
   from 2,646 templates
   — validate_node_minimal() + validate_node_operation()
   to pre-validate before building
   — validate_workflow() + validate_workflow_connections()

   • validate_workflow_expressions() post-build
     — n8n_create_workflow() to deploy live
     — n8n_validate_workflow() to confirm deployment

4. MCP Client Tool (official n8n node, no community packages)
   connects to the n8n-mcp Docker container on the same
   Docker network — all 42 tools exposed, bearer auth secured

5. Respond To Chat sends only the workflow URL and
   a credential checklist — never raw JSON

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🔧 WHAT YOU NEED
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

→ Self-hosted n8n (Docker)
→ n8n-mcp container on the same Docker network
(full docker-compose in sticky notes)
→ OpenRouter API key (or native Anthropic node)
→ Postgres for chat memory
→ n8n API key with workflow create permissions
→ Bearer token credential for n8n-mcp

No community nodes required. MCP Client Tool is an
official built-in node in n8n.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🛠️ NODES IN THIS WORKFLOW
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

→ Chat Trigger — public chat UI, auto session management
→ AI Agent (Tools Agent, v3.1) — 25 iterations, full prompt
→ MCP Client Tool (official) — all 42 tools, bearer auth
→ OpenRouter Chat Model — Claude Sonnet 4.5
→ Postgres Chat Memory — 10-message window per session
→ Respond To Chat — formatted reply with error fallback

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🔁 CUSTOMISATION
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

→ Swap to any LLM via OpenRouter or native Anthropic node
→ Adjust Postgres memory window (default: 10 messages)
→ Restrict the system prompt to specific workflow categories
→ Add a human approval step before deployment
→ Deploy n8n-mcp on Railway or any VPS for remote access

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
⚠️ SELF-HOSTED ONLY
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Requires self-hosted n8n. The n8n-mcp container must be
reachable from n8n over HTTP on the same Docker network.
Not compatible with n8n Cloud without exposing the MCP
server publicly.

Docker Compose.yml and .env files
https://docs.google.com/document/d/1qvkJnybhjeS3T2ySKsJ0MZF1rPZYcC1nbxBJjxC6Pqk/edit?usp=sharing

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📬 SUPPORT
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Email: [email protected]
LinkedIn: https://www.linkedin.com/in/salman-mehboob-pro/