Log inStart Free Trial
Meta Tech Provider

WhatsApp prompts for ChatGPT and Claude: 40 requests that run your WhatsApp

These 40 requests work in ChatGPT or Claude once the easyCoexistence connector is added for a WhatsApp Business number. They are grouped by job, from checking the number to templates, messages, automations, AI replies and troubleshooting, and each is mapped to the MCP tool the assistant calls, so you know what will happen before you approve it.

30+ tools

In the connector behind these requests, reachable from ChatGPT, Claude, Claude Code and Codex with one OAuth sign-in.

list_numbers

The first tool assistants call, because the tools that act on a number take the connection_id it returns.

24 hours

After the customer's last message, the limit for send_message; later, only an approved template can be sent.

Up to 24 hours

Meta's review time for a template submitted with create_template.

131047

Meta's error for a free-form message sent too long after the customer's last message.

1,380 minutes

Longest Wait step in an automation, which is 23 hours.

How do you use these requests?

Add the connector once, then type a request as written or in your own words. In ChatGPT on the web, type @ and pick easyCoexistence; in Claude, turn the connector on from the + menu, under Connectors. The assistant calls list_numbers first, because the tools that act on a number take the connection_id it returns, so name the number when you have more than one. Write phone numbers in international format, such as +1 415 555 0142, since the send tools take a number; for people who have already written to a number that comes through easyCoexistence, the assistant can look a name up with list_conversations.

Expect approvals. ChatGPT asks you to confirm the tool calls it treats as writes, and the connector marks its reading tools read-only, so only the tools that send or change something ask; Claude shows a tool approval request unless you have approved that tool permanently.

Which requests check the number and connect a new one?

Reading requests are the safest place to start: they change nothing, and they keep working while billing is paused, except the credentials, which carry a token. get_number_health combines a live snapshot from Meta with the last 15 events of the monitoring timeline, which is how an assistant explains a drop such as the 14-day rule. Connecting a number needs a browser, so the assistant hands you the link and the steps rather than doing it.

1. "Which WhatsApp numbers are connected to my account?"list_numbers: status, quality rating and messaging tier
2. "Is +1 415 555 0142 healthy?"get_number_health: live quality, status and throughput from Meta
3. "Is the WhatsApp Business app still linked to my number?"get_number_health: coexistence_link_ok, from Meta's is_on_biz_app
4. "What happened to this number this week?"get_number_health: the last 15 monitoring events
5. "How do I connect another number?"get_connect_link: the dashboard link and steps, Meta's window plus a QR code
6. "Why do numbers disconnect after 14 days, and how do I avoid it?"get_docs: answered from easyCoexistence's own documentation
7. "Give me the phone number ID and WABA ID for my n8n setup."get_api_credentials: IDs, send endpoint and your token when available

Which requests create and send templates?

Templates are how a business writes first, or writes again after 24 hours of silence from the customer, and on a Coexistence number they can only be sent through the API, never from the phone app. create_template submits to Meta in one of three categories, UTILITY, MARKETING or AUTHENTICATION, with numbered variables such as {{1}} and an example value for each. Meta's review can take up to 24 hours, and marketing templates need a payment method on your WhatsApp Business Account.

8. "List my approved templates and their languages."list_templates: name, status, category, language and quality
9. "Create a utility template in Portuguese saying an order shipped, with the order number as a variable."create_template: UTILITY, pt_BR, a body with {{1}} and an example
10. "Create a marketing template for new customers with the coupon BEMVINDO10."create_template: MARKETING; needs a payment method on the account
11. "Has Meta approved order_shipped yet?"get_template_status: APPROVED, PENDING or REJECTED
12. "order_shipped was rejected. Rewrite it and submit it under a new name."get_template_status, then create_template
13. "Send order_shipped to +1 415 555 0142 with order 1042."send_template: 1042 as the body parameter
14. "Which of my templates are marketing and which are utility?"list_templates: the category of each template

Which requests send messages?

send_message sends text only within 24 hours of that customer's last message to your number; later, Meta refuses it with error 131047, and only send_template works. Messages sent this way come from your business number and appear in the chat thread of the WhatsApp Business app. On a number whose messages come through easyCoexistence, the assistant can see when a customer last wrote with list_conversations and get_conversation; otherwise ask it to fall back to a template when the text is refused.

15. "Tell +1 415 555 0142 their table for 4 is booked for Sunday at 11."send_message: text, within 24 hours of their last message
16. "Message +1 415 555 0142; if Meta refuses it, send the follow_up template."send_message, then send_template after error 131047
17. "Send appointment_reminder to the three customers booked for tomorrow: [numbers]."send_template, once per number
18. "Send a test message to my own phone, +1 415 555 0177."send_message; write to the business number from that phone first
19. "Reply to Ana at +1 415 555 0142 with our address and opening hours."send_message: the assistant needs the number, not only the name

Which requests build automations?

An automation is one trigger and a flow of steps, up to 30 in a line and 60 in all, with branches (condition, ai_decide), questions that wait for the answer (ask) and AI agents (agent), and it answers customers without the assistant being open. create_automation switches it on by default, so say "create it switched off" when you want to review it first; test_automation runs the saved steps on a sample message without sending anything. Only one automation answers a given message: keyword automations first, then first message, then after hours, then the catch-all.

20. "When someone writes to us for the first time, send the welcome_coupon template."create_automation: First message, Send a template
21. "Outside 9:00 to 18:00, Monday to Friday, São Paulo time, reply that we open at 9."create_automation: After hours, America/Sao_Paulo, days 1 to 5
22. "If a message contains price or menu, reply with our price list."create_automation: Has keyword, Contains, Reply with text
23. "Only when the whole message is PIX, reply with our payment key."create_automation: Has keyword, Is exactly
24. "Two hours after a first message, send the follow_up template unless I have answered."create_automation: First message, Wait 120 minutes, Send a template
25. "Pass messages from this number to my n8n webhook and send back its reply."create_automation: Call my webhook, reply within 20 seconds
26. "Also forward this number's messages to https://example.com/hooks/wa."set_inbound_mode: forward_url, signing secret shown once
27. "List my automations and tell me which ones are off."list_automations: trigger, steps and on or off
28. "Turn off the away message for the holidays."update_automation: enabled false; ask again to turn it back on
29. "Delete the old welcome automation."delete_automation: its run history goes with it

Which requests set up AI replies?

AI replies run on your own key, from Anthropic, OpenAI or OpenRouter, saved by you under AI keys in the dashboard, since no tool accepts a key. Without one, create_automation refuses the AI step, and list_automations shows which providers have a key. An AI step has a provider, a model id, instructions of up to 20,000 characters, how many earlier messages it reads (0 to 20, default 10) and the longest reply in tokens (50 to 2,000, default 600). For open conversation, create_agent builds an AI agent with instructions and knowledge, add_agent_knowledge teaches it, test_agent tries it, and an agent step attaches it to an automation.

30. "Create an AI assistant that answers questions about our menu, with my OpenAI key, in short friendly messages."create_automation: Reply with AI, provider openai
31. "Which AI keys have I saved?"list_automations: provider and status, never the key
32. "Make the AI read the last 20 messages and keep replies under 300 tokens."update_automation: history 20, max_tokens 300
33. "After I answer a chat from my phone, keep the AI quiet there for 8 hours."update_automation: takeover_hours 8
34. "Allow at most 5 AI replies per chat per hour."update_automation: ai_max_per_hour 5
35. "Test the AI assistant with: Do you have gluten-free bread?"test_automation: the model really answers, nothing is sent

Which requests find out why something did not reply?

Run logs answer most "why" questions. list_automation_runs returns the last 30 days of an automation's runs, newest first, with the status, the skip reason and the tokens used; get_automation_run opens one run with each step's output, error and timing. The usual skip reasons are human_active (you answered from the phone recently), window_closed (more than 24 hours since the customer wrote), ai_limit and superseded.

36. "Why didn't the welcome automation reply to the last customer?"list_automation_runs: status and skip_reason
37. "Show me the steps of the last failed run, with the error."get_automation_run: output, error and timing per step
38. "Why is my automation not running at all?"list_automations shows on, off or broken; a direct number needs set_inbound_mode managed
39. "How many AI tokens did the assistant use today?"list_automation_runs: tokens in and out per run
40. "My number shows disconnected. What should I do?"get_number_health and get_docs: the reason, then how to reconnect

Which requests will not work, and why?

Some requests sound reasonable but have no tool behind them, and a well-informed assistant will say so instead of improvising:

  • "Read my last 50 WhatsApp messages": no tool can read a number whose messages go straight to your own system, because they never pass through easyCoexistence; on a managed number, get_conversation reads a contact's chat for the retention you set.
  • "Send this to all my contacts": there is no broadcast tool, and business-initiated messages need an approved template and the person's opt-in, sent one at a time with send_template.
  • "Here is my OpenAI key": keys are saved in the dashboard under AI keys, never through the chat.
  • "Post in my customers' group": group chats are not available through the API on a Coexistence number.
  • "Connect my number for me": Meta's window and the QR code need you, in a browser.
  • "Call this customer": voice and video calls keep working in the app, but the Cloud API does not carry them on a Coexistence number.
  • "Transcribe this voice note": automations write incoming voice notes out as "[voice message]" and the words, but only with an OpenAI or OpenRouter key saved, and no tool transcribes a file you give it.

Common mistakes

  • Asking for a send by name only. The send tools take a number, so give it in international format, or have the assistant find the person with list_conversations first; the tool keeps only the digits.
  • Using Contains for a short keyword. "hi" with Contains also fires on "this" and "which"; use Starts with or Is exactly for short words.
  • Approving create_automation without reading it. It is switched on by default and starts answering the moment it is saved, so check the trigger and the text before you confirm.
  • Pasting an AI key to fix a missing-key error. No tool takes keys: save the key under AI keys in the dashboard, then repeat the request.
Doing this with EasyCoexistence

easyCoexistence is the MCP connector behind these requests: more than 30 tools on the WhatsApp Business number you already use, through Meta's Cloud API and Coexistence, with one OAuth sign-in in ChatGPT, Claude, Claude Code or Codex. The app stays on your phone. US$ 9 down to US$ 0.75 per number per month, falling as you add numbers, with the first 3 days free.

easyCoexistence is a multichannel API and MCP server that connects AI agents such as Claude and ChatGPT to WhatsApp, Instagram, TikTok, Facebook and more. On WhatsApp it uses Coexistence: the business keeps using its number in the WhatsApp Business app while the AI works through Meta's official API.

Frequently asked questions

Do I have to use these exact words?

No. The assistant picks the tool from plain words. What saves a follow-up question is naming the number, the template, the timezone and the exact text to send.

Can I write the requests in Portuguese or Spanish?

Yes. ChatGPT and Claude call the same tools in any language. Keywords match letter for letter apart from capitals, so add both cardápio and cardapio if customers type either.

Can the assistant undo what it did?

Partly. It can turn an automation off, change it or delete it, and change where messages go, but a sent message cannot be recalled through the connector, and there is no tool to delete a template.

Do these requests work in Claude Code and Codex?

Yes. Both reach the same tools: claude mcp add --transport http easycoexistence --scope user https://easycoexistence.com/api/mcp, or codex mcp add easycoexistence --url https://easycoexistence.com/api/mcp followed by codex mcp login easycoexistence.

Keep reading

Ready to get started?

Set up WhatsApp Coexistence in minutes, not months. The app keeps working on the phone.

Start Free Trial

Verified on