Let Claude answer your WhatsApp customers
Claude can answer WhatsApp Business customers through Meta's official API: an automation sends the incoming message, your instructions and the recent chat to Anthropic's API on your own key, and Claude's reply goes out from your number. You create the key in the Claude Console, pick a Claude model id, and keep using the WhatsApp Business app on your phone.
easyCoexistence runs the automation that puts Claude on your WhatsApp Business number: your Anthropic key, your Claude model and your Instructions, with the chat history, reply caps and the pause when you answer from the phone built in.
How a Claude API key starts; the Console shows it in full only once, when it is created.
A paid Claude chat plan and the Claude API are billed separately, though Max, Ultra and Team include some monthly API credits.
Anthropic's answer when a key not scoped to a workspace calls the API without a workspace ID header.
Anthropic's billing_error: a billing or payment problem on the account, apart from 401 for a bad or expired key.
Thinking tokens count toward it, so a long thinking pass can use the budget before the reply text finishes.
The longest a Reply with AI step waits for the model before that run fails.
How does Claude reply to a WhatsApp customer?
Through a Reply with AI step whose Provider is Anthropic. A customer writes, Meta delivers the message to easyCoexistence over the Cloud API, the matching automation calls Anthropic's Messages API with your key and the model you chose, and Claude's text is sent back on the same chat from your WhatsApp Business number. The reply also shows in the app on your phone.
The Claude app itself plays no part in this, and nobody has to keep a chat open. Claude can still build the automation for you through the easyCoexistence connector in Claude, which calls create_automation and then test_automation, but answering customers runs on the API key you saved.
Which Anthropic key works, and how do you create one?
A Claude API key from the Claude Console, not a claude.ai login. Open the API keys page at platform.claude.com/settings/keys, click Create key, name it, choose an expiration and a linked account, and copy the key that starts with sk-ant-: the Console never shows it again. Add credits in the Console's billing settings before customers depend on it.
Two choices at creation decide whether replies keep working. Scope the key to one workspace: the automation sends the key alone, with no workspace header, and Anthropic refuses an unscoped key's requests with a 400 error. Pick a long expiration or Never, because an expired key returns 401 and pauses your Claude automations. Then, on the Automations page, open AI keys, click Add key next to Anthropic, paste it and click Test and save. The check reads Anthropic's model list, spends no tokens, and fills the Model picker.
Which Claude model should answer?
The Model field lists the models Anthropic returns for your key, and you can also type an id. Anthropic describes its current lineup by job and speed, which maps well to chat replies: a customer waiting on WhatsApp notices seconds. Each Claude id is a pinned snapshot, so a model keeps behaving the same until Anthropic retires it; after retirement Anthropic answers 404 and those runs fail until you pick a newer id.
Why does a Claude reply need more room in Longest reply?
Because current Claude models think before they answer, and that thinking uses the same token budget as the reply. Anthropic lists thinking as on at all times for Claude Opus 5.5 and on by default for Claude Sonnet 5.5 and Claude Haiku 5.5. The Reply with AI step sends no thinking or effort setting, so the model's own default applies, and Longest reply (tokens) becomes the cap for thinking plus text.
At the default of 600, a question that makes the model think at length can leave the reply cut off mid-sentence, or empty, in which case the run fails with "The model returned an empty reply" and you receive an email. For a thinking model, set Longest reply between 1,000 and 2,000 and keep reply length under control in the Instructions instead, for example "answer in 1 to 3 short sentences". Thinking tokens appear in the out count of each run and are billed by Anthropic.
What does Claude receive from the chat?
A system prompt and a short conversation. The system prompt is your Instructions, up to 20,000 characters, with {{first_name}}, {{name}}, {{message}} and {{phone}} filled in, followed by five lines easyCoexistence adds: write a short plain-text chat message in the customer's language, say a person will answer soon when asked or stuck, never invent prices, stock, dates or policies, and say it is the business's automated assistant when asked, which the first reply after a quiet day does unprompted.
The conversation holds up to 20 earlier messages of the chat, kept for the days you choose in Conversations (30, 90 or 365), 10 by default, ending with the new one. Customer messages become user turns and replies from your number, typed on the phone or sent by an automation, become assistant turns. Consecutive messages from the same side are joined into one turn, and business messages at the very start are dropped, since some providers refuse a conversation that opens with the assistant. Photos arrive as a placeholder such as "[image]". A voice note is written out as "[voice message]" followed by the words only when you also have an OpenAI or OpenRouter key, since Anthropic cannot write out speech; without one it stays "[voice message]".
What happens when Anthropic returns an error?
It depends on the status code, and the run log keeps Anthropic's message either way. Codes that point at the account pause every automation on your account that uses Anthropic, because no reply can succeed until you fix them. Codes that point at one request fail only that run: rate limits and overloads get one quick retry first, and a failed run emails you, at most once per automation every 6 hours, so you can answer that customer yourself.
Can Claude set this up for you?
Yes, once your Anthropic key is saved in the dashboard, since no tool accepts a key. In Claude with the easyCoexistence connector turned on, describe the assistant you want and ask for it switched off so you can test first. Claude calls list_numbers, then create_automation with provider anthropic, then test_automation with a sample question. Without a saved key, create_automation refuses the step with "save your anthropic API key on the Automations page first".
{
"connection_id": "<from list_numbers>",
"name": "Claude answers questions",
"enabled": false,
"trigger": { "type": "message_received", "match": "any" },
"steps": [
{
"type": "ai_reply",
"provider": "anthropic",
"model": "claude-sonnet-5-5",
"prompt": "You are the assistant of Bloom Cafe. Open daily 8:00 to 18:00. ...",
"history": 10,
"max_tokens": 1500
}
],
"settings": { "takeover_hours": 4, "ai_max_per_hour": 10 }
}How to set it up
- Create a Claude API keyIn the Claude Console, open API keys, click Create key, scope it to one workspace, choose a long expiration or Never, and copy the sk-ant- key.
- Add creditsAdd credits or a payment method in the Console's billing settings, so Anthropic accepts the replies.
- Save it under AI keysOn the Automations page, open AI keys, click Add key next to Anthropic, paste the key and click Test and save. It then shows as Added, ending in its last four characters.
- Add a Reply with AI stepClick New automation, then Start from scratch, delete the empty Reply with text step and add a Reply with AI step with the + under Say something. Set Provider to Anthropic, pick a Claude model and write your Instructions.
- Give thinking roomRaise Longest reply (tokens) to between 1,000 and 2,000, and ask for short answers in the Instructions.
- Save it Off, test, then turn it onFlip the switch to Off, click Save, then Test and Run test with real questions. Claude really answers and nothing is sent. Flip it On and save.
Common mistakes
- Creating the key without a workspace. The automation sends no workspace header, so Anthropic answers 400 on every reply, even if the key passed Test and save.
- Picking a 7-day or 30-day expiration. When the key expires, Anthropic returns 401, every Claude automation pauses, and nothing answers until a new key is saved.
- Leaving Longest reply at 600 on a model that thinks. Thinking shares that budget, so long questions can end in a cut-off or empty reply.
- Typing OpenRouter's form of a Claude id, such as anthropic/claude-sonnet-5.5, under Provider Anthropic. Anthropic's ids use dashes, such as claude-sonnet-5-5, and an unknown id fails with 404.
We never charge for AI or add a markup, and automations cost nothing extra. 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
Can I use my Claude Pro or Max plan instead of an API key?
No. Replies call the Claude API, which needs a key from the Claude Console and is billed there. Anthropic bills the API separately from chat plans; Max, Ultra and Team include some monthly API credits that the key's organization can use.
Can I reach Claude through OpenRouter instead?
Yes. Choose Provider OpenRouter and type a Claude id from OpenRouter's catalog, such as anthropic/claude-sonnet-5.5. The tokens are then billed through your OpenRouter credits instead of your Anthropic account.
Can Claude and an OpenAI model answer on the same number?
Yes, in different automations. Only one automation answers a given message, so you could run Claude in the catch-all and an OpenAI model in a keyword automation, each on its own saved key.
Does Claude still answer after 24 hours without a customer message?
No. Free-form replies, Claude's included, go out only within 24 hours of the customer's last message, as Meta requires; later runs are skipped as window_closed, and only an approved template can reach that customer.
Keep reading
Ready to get started?
Set up WhatsApp Coexistence in minutes, not months. The app keeps working on the phone.
Start Free TrialVerified on