n8n Complete Tutorial: Workflow Automation, Data Pipelines and AI Agents

What n8n is

n8n ("n-eight-n", from nodemation) connects apps and data with visual workflows. Each workflow starts with a trigger (a form is submitted, a schedule fires, a webhook is called) and passes data through nodes that read, transform and send it. Compared with Zapier and Make, n8n can be self-hosted (your data stays on your server), lets you write code in any step, and has strong AI-agent nodes.

Getting n8n

The editor and the core ideas

Items: the most important idea

Every node receives and outputs an array of items, each { "json": { … } }. If a Google Sheets node returns 25 rows, the next Slack node runs 25 times, once per item, unless you aggregate first.

import json
items = [{"json": {"name": "Asha", "email": "[email protected]", "city": "Pune", "score": 82}},
         {"json": {"name": "Ravi", "email": "[email protected]", "city": "Delhi", "score": 45}},
         {"json": {"name": "Meera", "email": "[email protected]", "city": "Pune", "score": 91}}]
# IF node: score >= 60 goes to the "true" branch
true_branch = [i for i in items if i["json"]["score"] >= 60]
# Edit Fields (Set) node: keep and add fields
shaped = [{"json": {"lead": i["json"]["name"], "email": i["json"]["email"], "priority": "hot" if i["json"]["score"] >= 90 else "warm"}} for i in true_branch]
# Aggregate node: many items -> one item with a list
aggregated = {"json": {"leads": [s["json"]["lead"] for s in shaped], "count": len(shaped)}}
print(json.dumps(shaped, indent=1)); print(aggregated)

[
 {
  "json": {
   "lead": "Asha",
   "email": "[email protected]",
   "priority": "warm"
  }
 },
 {
  "json": {
   "lead": "Meera",
   "email": "[email protected]",
   "priority": "hot"
  }
 }
]
{'json': {'leads': ['Asha', 'Meera'], 'count': 2}}

Core nodes you will use every day

Code node example (JavaScript, "Run once for all items")

// Clean phone numbers and drop duplicates by email
const seen = new Set();
return $input.all()
  .filter(item => { const e = item.json.email.toLowerCase().trim(); if (seen.has(e)) return false; seen.add(e); return true; })
  .map(item => ({ json: { ...item.json, phone: String(item.json.phone || '').replace(/\D/g, '').slice(-10) } }));

The Python version of the same idea (Code node → Language: Python) uses _input.all() and returns a list of dicts with a json key.

HTTP Request: calling an API

# What an HTTP Request node sends and how you map the answer back into items
request = {"method": "GET", "url": "https://api.example.com/v1/orders", "qs": {"status": "paid", "limit": 2},
           "headers": {"Authorization": "Bearer {{ $credentials.apiKey }}"}}
response = {"data": [{"id": 101, "amount": 2499, "city": "Pune"}, {"id": 102, "amount": 899, "city": "Delhi"}], "next": None}
items = [{"json": o} for o in response["data"]]          # Split Out on "data"
print(request["method"], request["url"], request["qs"]); print(items)

GET https://api.example.com/v1/orders {'status': 'paid', 'limit': 2}
[{'json': {'id': 101, 'amount': 2499, 'city': 'Pune'}}, {'json': {'id': 102, 'amount': 899, 'city': 'Delhi'}}]

Tips: use Authentication → Predefined credential type when n8n has the app, otherwise Generic credential (header auth, OAuth2); turn on Pagination for APIs with next links; add Retry on fail (Settings) for flaky APIs.

Reliability: errors, retries and scheduling

• Settings per node: Retry on fail (tries and wait), Continue on fail / error output branch, Always output data.
• Error workflow: Workflow settings → Error workflow → pick a workflow that starts with Error Trigger and sends you a Slack/WhatsApp/email alert with the failed node and message.
• Schedule Trigger: every X minutes, daily at 09:00 IST (set the workflow time zone to Asia/Kolkata), or a cron expression 0 9 * * 1-5.
• Pin data while building: freeze a node's output so you do not call the API on every test.
• Environment safety: keep secrets in credentials, never in Code nodes; restrict who can edit workflows; use separate dev and prod instances for important flows.

AI agents in n8n

The AI Agent node runs a loop: a chat model reads the goal, decides which tool to call, reads the result and continues until it can answer.

Other AI building blocks: Basic LLM Chain, Text Classifier, Information Extractor (pull fields from emails or PDFs into JSON), Summarization Chain, Sentiment Analysis, Embeddings + Vector Store nodes (Pinecone, Qdrant, Supabase, PGVector, in-memory) for RAG, and the Chat Trigger for a hosted chat window.

n8n and MCP

• MCP Client Tool: lets an n8n agent use tools from any MCP server.
• MCP Server Trigger: exposes your n8n workflows as MCP tools, so Claude or other MCP clients can call them (for example "create a lead in our CRM").

Four complete projects

1. Lead capture to Google Sheets and Slack

n8n Form Trigger (name, email, phone, course) → Edit Fields (clean phone, add timestamp) → Google Sheets: Append row → IF course = "Data Engineer" → Slack: send message to #sales (true) / Gmail: send brochure (false).

2. Daily sales report

Schedule Trigger 09:00 IST → Postgres/HTTP Request: yesterday's orders → Summarize: sum sales by region → Code: build an HTML table → Gmail: send to managers + Slack summary.

import pandas as pd
orders = pd.DataFrame({"region": ["South", "West", "South", "North", "East", "West"], "amount": [24990, 8999, 15500, 4300, 12000, 30500]})
summary = orders.groupby("region").amount.agg(["count", "sum"]).sort_values("sum", ascending=False)
html = summary.to_html()            # the Code node builds the same table for the email body
print(summary); print("HTML length for the email:", len(html), "characters")

        count    sum
region              
South       2  40490
West        2  39499
East        1  12000
North       1   4300
HTML length for the email: 571 characters

3. Email triage AI agent

Gmail Trigger (new email) → Text Classifier (Sales lead / Support / Invoice / Spam) → Switch → Sales: Information Extractor (name, company, budget) → HubSpot/Sheets + Slack ping; Support: AI Agent drafts a reply using a Vector Store tool over your FAQ and saves it as a Gmail draft for a human to send; Spam: archive.

4. Document Q&A bot (RAG)

Indexing workflow: Google Drive Trigger (new PDF) → Default Data Loader + Recursive Text Splitter → Embeddings → Vector Store: insert. Chat workflow: Chat Trigger → AI Agent with the vector store as a tool, a chat model and window memory → answer with sources. This is the RAG pattern from the GenAI track, built without code.

Summary

Learn more (official and free)

• n8n documentation
• n8n free courses, Level 1 and Level 2
• n8n workflow templates (thousands of ready examples)
• n8n community forum
• Advanced AI in n8n (agents, RAG, MCP)