n8n + Claude API: Build Your First AI Automation in India (UPI Billing, No Card Required)

n8n has become the go-to automation tool for Indian freelancers and indie developers. It's visual, self-hostable, and the node library now covers virtually every service you'd want to connect. The pattern that's working for dozens of Indian developers I know: build automations with n8n, power the AI parts with Claude, charge clients for the outcome.

The historic friction point: Claude's API required an international credit card. Most Indian developers don't have one — or don't want to use a personal card for a business tool.

That's solved now. AICredits.in gives you Claude API access with UPI billing. Add ₹200 via GPay, get an API key, you're running in 5 minutes.

This tutorial walks you through a complete WhatsApp business FAQ bot — one of the most requested automations for Indian small businesses. By the end you'll have a working n8n workflow that answers customer questions automatically, 24/7, in under ₹15/day.

What you'll build

Option A (this tutorial): WhatsApp business FAQ bot A customer messages your WhatsApp number with a question. Claude reads the message, checks your FAQ knowledge base, and replies with a relevant answer. If it can't answer, it hands off to a human agent.

Option B: Gmail inbox triage New emails arrive → Claude categorizes them (sales lead / support request / spam / newsletter) and creates tasks in Notion or assigns to team members in Slack.

Option C: Invoice data extractor PDF invoices arrive in a shared email → Claude extracts vendor name, invoice number, amount, due date → creates rows in Google Sheets automatically.

All three use the same core n8n + Claude pattern. Once you've built Option A, you can wire up B and C in under an hour.

Setup

Step 1: Get your AICredits.in API key

This takes 5 minutes.

1. Go to aicredits.in and create an account
2. Click "Add Credits" → choose ₹200 (enough for ~400,000 tokens — plenty to test this workflow)
3. Pay via UPI/GPay
4. Go to API Keys → Create Key → copy it somewhere safe

You'll use this key in the n8n setup below. The base URL you'll need: https://api.aicredits.in/v1

Step 2: Set up n8n

Two options:

Cloud n8n (easiest to start): Go to n8n.io and create a free account. The free tier gives you 5 active workflows — enough to build and test this.

Self-hosted with Docker (better for production):

docker run -it --rm \
  --name n8n \
  -p 5678:5678 \
  -v ~/.n8n:/home/node/.n8n \
  n8nio/n8n

Open http://localhost:5678 in your browser. Create an account when prompted.

Self-hosting is worth it if you're building client automations — you control the data, and there's no per-workflow limit.

Step 3: Configure the OpenAI node to use AICredits

Here's the key insight: n8n's OpenAI node supports custom base URLs. AICredits.in exposes an OpenAI-compatible API, so you configure the node as if it's OpenAI but point it at AICredits.

In n8n:

1. Go to Settings → Credentials → New Credential
2. Choose OpenAI API
3. In the API Key field: paste your AICredits.in API key
4. In the Base URL field: https://api.aicredits.in/v1
5. Save as "AICredits - Claude"

When you add an OpenAI Chat Model node to any workflow, select this credential. For the model name, use: anthropic/claude-sonnet-4-6

That's it. Every OpenAI node in n8n now routes through Claude via AICredits.

Build the WhatsApp FAQ bot

The workflow overview

[WhatsApp Webhook] → [Extract Message] → [Claude Chat] → [Check Confidence] → [WhatsApp Reply]
                                                               ↓
                                                    [Low confidence → Human handoff]

Step-by-step n8n nodes

Node 1: Webhook (WhatsApp trigger)

Add a Webhook node. This receives incoming WhatsApp messages via Twilio or the WhatsApp Business API.

• Method: POST
• Path: /whatsapp-incoming
• Response mode: Respond to Webhook

Copy the webhook URL — you'll paste this into your WhatsApp/Twilio setup.

If you're using Twilio WhatsApp sandbox (easiest to test):

• Twilio Console → WhatsApp → Sandbox
• Set "When a message comes in" to your n8n webhook URL

Node 2: Set node — extract message

Add a Set node after the webhook. Extract the customer message:

customer_message: {{ $json.Body }}
from_number: {{ $json.From }}

(Twilio sends the message body in Body and sender in From. Other providers use different field names — check their webhook docs.)

Node 3: OpenAI Chat Model node — Claude

Add an AI Agent node (n8n's LangChain-based agent node). Configure it with:

• Model: OpenAI Chat Model (using your AICredits credential)
• Model name: anthropic/claude-sonnet-4-6

Set the system prompt:

You are a customer support assistant for [BUSINESS_NAME], a [TYPE_OF_BUSINESS] based in India.

Your job is to answer customer questions based on the FAQ below. 

IMPORTANT RULES:
1. Only answer based on the FAQ provided. Do not make up information.
2. If you cannot find a relevant answer in the FAQ, respond with exactly: "HUMAN_HANDOFF_REQUIRED"
3. Keep responses under 200 words — this is WhatsApp, not email.
4. Write in simple, clear English. If the customer wrote in Hindi, respond in Hindi.
5. Be helpful and friendly, but don't add unnecessary pleasantries.
6. For pricing questions, always direct to: [PRICING_URL_OR_CONTACT]

FAQ:
[PASTE_YOUR_FAQ_HERE — one Q&A per section, formatted as Q: / A:]

Current date: {{ $now.toISO() }}

For the user message input: {{ $json.customer_message }}

Node 4: IF node — confidence check

Add an IF node that checks if the Claude response contains "HUMAN_HANDOFF_REQUIRED":

• Condition: Output text contains "HUMAN_HANDOFF_REQUIRED"
• True branch → human handoff notification
• False branch → WhatsApp reply

Node 5a: WhatsApp reply (automated)

Add an HTTP Request node to send the reply:

For Twilio:

• Method: POST
• URL: https://api.twilio.com/2010-04-01/Accounts/[YOUR_ACCOUNT_SID]/Messages.json
• Authentication: Basic (your Twilio credentials)
• Body: From=[TWILIO_WHATSAPP_NUMBER]&To={{ $json.from_number }}&Body={{ $json.output }}

Node 5b: Human handoff (when Claude can't answer)

Add a notification step — Slack message, email, or create a ticket:

Example Slack notification:

🔔 WhatsApp message needs human response
From: {{ $json.from_number }}
Message: {{ $json.customer_message }}
React to this message to claim it.

The Claude system prompt for business bots — key patterns

A few patterns that make business FAQ bots reliable:

1. Format the FAQ for scanning, not reading:

Q: What are your delivery charges?
A: Delivery is free for orders above ₹499. Below ₹499, we charge ₹49 flat.

Q: How long does delivery take?
A: Metro cities (Delhi, Mumbai, Bengaluru, Chennai, Hyderabad): 2-3 business days. Other cities: 4-6 business days.

2. Add boundary instructions explicitly: The line "Only answer based on the FAQ provided" sounds obvious, but without it, Claude will helpfully extrapolate and sometimes get things wrong.

3. Handle the language question: Indian customers write in English, Hinglish, Hindi, or regional languages. Add: "If the customer wrote in [LANGUAGE], respond in [LANGUAGE]" for the languages your business serves.

4. Product catalogue as context: If you have a product catalogue, paste the top 20-30 items (with prices and key attributes) into the system prompt. Claude will answer product questions directly rather than deflecting every time.

Cost estimate

Let's make this concrete for a typical Indian SME use case.

Assumptions:

• 100 WhatsApp messages per day
• Average message: ~50 words (250 tokens)
• Average Claude response: ~100 words (500 tokens)
• Input tokens per request: ~1,500 (system prompt + message) = average with FAQ
• Output tokens per request: ~500

Daily token usage: 100 × (1,500 + 500) = 200,000 tokens

Claude Sonnet pricing via AICredits.in (approximate): ₹0.21 per 1,000 input tokens, ₹0.84 per 1,000 output tokens

Daily cost: (100 × 1,500 × 0.00021) + (100 × 500 × 0.00084) = ₹3.15 + ₹4.20 = ₹7.35/day

Monthly: ~₹220/month for 24/7 AI customer support handling 100 messages/day.

For context, a WhatsApp Business API account costs around ₹800–1,500/month depending on message volume. The Claude cost is a fraction of the infrastructure cost.

Try it now with AICredits.in

Access Claude, GPT-4o, Gemini, and 300+ models with UPI payment in ₹. No international card needed. ₹200 in credits is enough to build and test this entire workflow. Create free account →

What to build next

Once you've got the WhatsApp FAQ bot running, these are the natural next workflows:

Gmail inbox triage: Webhook triggers on new email → Claude categorizes (sales lead / support / partnership / spam) → routes to the right team member via Slack or creates tasks in Notion. The system prompt pattern is identical — just swap the FAQ for routing rules.

Lead qualifier: Form submission from your website → Claude asks 3 qualifying questions via WhatsApp/email → scores the lead (hot/warm/cold) based on answers → notifies sales team with a summary. This is the automation that converts form leads into qualified conversations without human effort.

Invoice extractor from PDFs: Email arrives with PDF attachment → n8n's PDF extraction node pulls text → Claude extracts vendor, invoice number, amount, GST details, due date → creates row in Google Sheets. Indian businesses dealing with 50+ invoices/month find this saves 2–3 hours weekly.

All three follow the same n8n + Claude structure. The only thing that changes is the system prompt and the downstream actions.

Next steps

• How to build AI calling agents with n8n — extend this to voice-based customer support
• n8n AI agent system prompt templates — more system prompt patterns for different use cases
• Function calling explained — understanding how Claude uses tools in agentic workflows