How to send WhatsApp messages from Claude or ChatGPT

Claude and ChatGPT can do far more than write your WhatsApp messages — with a connector they can actually send them: submit templates to Meta, check who's inside the reply window, send to a spreadsheet of customers, and tell you how it went. This guide shows how to set that up and what it looks like in practice.

How it works

Both assistants support custom connectors built on the Model Context Protocol (MCP), an open standard for giving AI assistants tools. SendFromChat is an MCP server in front of Meta's official WhatsApp Cloud API. Once connected, your assistant sees tools like create_template, check_recipient, send_message, upload_contacts and send_campaign, and calls them when you ask for something.

You stay in control: the assistant shows you what it's about to send and asks before bulk sends go out.

What you need

1. A WhatsApp Business number on the Cloud API — setup guide (a free Meta test number is fine to start).
2. A SendFromChat workspace (30 days free, no card) with that number connected.
3. Claude: a plan that supports custom connectors. ChatGPT: a paid plan with Developer mode connectors enabled.

In the SendFromChat dashboard, the Overview page has a Connect your AI assistant card — generate your personal connector URL there. It contains a private key, so treat it like a password.

Connect Claude

1. In Claude (web or desktop), open Customize → Connectors (or Settings → Connectors).
2. Click + → Add custom connector.
3. Name it SendFromChat and paste your connector URL.
4. Click Add. No sign-in step is needed — the key is in the URL.
5. Start a new chat and say: "Help me get set up." Claude checks your setup status and walks you through anything left.

Using Claude Code in a terminal instead? One command:

claude mcp add --transport http sendfromchat https://sendfromchat.com/api/mcp/YOUR_KEY

Connect ChatGPT

1. Open Settings → Apps & Connectors and make sure Developer mode is on.
2. Create a new connector, name it SendFromChat, and paste your connector URL.
3. Set authentication to None — the URL already carries your key — and save.
4. In a new chat, enable the connector and say "Help me get set up."

Things to ask

Reply to a customer

Is +44 7700 900418 still inside the 24-hour window? If so, tell her the order ships tomorrow.

The assistant calls check_recipient first. If the customer messaged in the last 24 hours, it sends a free, free-form reply. If not, it suggests an approved template — and only drafts a new one if none fits. (Why that matters.)

Create a template

Draft a template telling customers their order has shipped, with their name and a tracking link, and submit it.

It writes the template with {{1}}/{{2}} variables and sample values, checks Meta's formatting rules, submits it, and can check back until it's approved.

Send to a list

Here's customers.xlsx — send the summer_sale template to everyone, using the Name column for {{1}}.

It previews the sheet (finding the phone column even if it's called "Mobile" or "WA No."), shows you rendered messages for real contacts and how many go out today, then queues the send after you confirm.

Automate

Set an away message for outside 9 to 6 New York time, and make MENU start a menu with Pricing, Support and Talk to a person.

Check results

How did yesterday's broadcast do? Who didn't get it and why?

Tips

• Start with a test number and your own phone as the recipient.
• Be specific about audiences ("everyone in the April list who hasn't opted out").
• Let it confirm — good assistants will show a preview before sending; keep that habit for bulk sends.
• Everything the assistant does is also visible in the SendFromChat dashboard — inbox, bulk sends, flows and sequences.

Prefer Cursor, VS Code, Gemini CLI or another tool? See using a WhatsApp MCP server in Cursor, VS Code and Gemini CLI.