איך בונים סוכן AI לוואטסאפ עם n8n ו-Claude ב-10 שלבים

Build a WhatsApp AI auto-reply workflow in my n8n instance, using my n8n MCP connection.

FLOW
1. WhatsApp Trigger, "messages" updates only.
2. Ignore status updates (sent / delivered / read) so it never loops.
3. Extract the customer's phone number and message text.
4. If the message is not text, reply: "תודה על ההודעה! כרגע אני יכול לקרוא רק טקסט. אפשר לכתוב לי את השאלה?"
5. AI Agent node + Anthropic Chat Model, with the business info and rules below as the system message.
6. Memory per customer phone number, last 10 messages.
7. Send the reply with the WhatsApp Business Cloud node, using my Phone Number ID.

BUSINESS INFO
- Business name: [שם העסק]
- What we do: [משפט או שניים]
- Products / services: [רשימה]
- Pricing: [מחירים, או "don't share prices, offer a call instead"]
- Hours: [ימים ושעות]
- Location: [כתובת או אזור]
- Website: [קישור]
- Policies: [משלוחים, החזרים, ביטולים]
- How to book / buy: [קישור או טלפון]
- Reply in the customer's language (usually Hebrew).

RULES
- Use only the info above. Never invent prices, policies or promises.
- 1-3 short sentences, like a person on WhatsApp. At most one emoji.
- Not sure? Say a teammate will get back to them.
- Never claim you did something you can't do (booked, refunded, sent a file).
- If the customer asks for a human or is upset, say a teammate will reply soon.

BUILD
- Reuse my existing n8n workflows and credentials where possible.
- If you can't create workflows through MCP, give me the full workflow JSON to paste.
- Use n8n credentials only. Never put keys or tokens in nodes or code.
- Name it "WhatsApp AI Agent" and leave it inactive.

When done, explain in simple Hebrew: what you built, which credentials to add, where to paste each one, and how to test.