WhatsApp API endpoints: the reference you can hold in your head

Base URL https://us.api-wa.me, pattern /{key}/..., where {key} is your instance key and also your authentication — there is no auth header to set. From there the surface splits into five groups: instance, messages, groups, contacts and webhooks.

That is genuinely the whole mental model. This page lists the routes so you can skim them in one sitting; the exhaustive parameter-by-parameter reference lives in /docs.

Ground rules

• Base URL: https://us.api-wa.me
• Auth: the instance key sits in the path — https://us.api-wa.me/YOUR_KEY/.... No header.
• Format: JSON request bodies; phone numbers in international format, digits only.

Because the key is the credential, the full URL is a secret. Keep it server-side, out of logs, and out of screenshots.

Instance and connection

Sending messages

The smallest thing that works:

curl -X POST "https://us.api-wa.me/YOUR_KEY/message/text" \
  -H "Content-Type: application/json" \
  -d '{ "to": "14155550132", "text": "Hello!" }'

Groups and contacts

That last one is worth a note: checking registration before a send is the cheapest way to keep a list clean, and on the official API it is the difference between a delivered template and one you paid for that went nowhere.

Webhooks: receiving events

Incoming messages arrive by webhook. Register them with PUT /{key}/instance:

curl -X PUT "https://us.api-wa.me/YOUR_KEY/instance" \
  -H "Content-Type: application/json" \
  -d '{
    "allowWebhook": true,
    "allowNumber": "all",
    "webhookMessage": "https://your-site.com/webhook",
    "webhookConnection": "https://your-site.com/webhook"
  }'

You can point each event type at a different URL — messages, connection changes, QR code, groups, your own outbound messages, history — and restrict which numbers fire them with allowNumber. Separate URLs are worth using: connection events and message events have very different failure modes, and mixing them into one handler is how a reconnect storm ends up in the same retry queue as customer replies.

Let an assistant do the integration

Two things make this API unusually easy to hand to a model:

• Every page has a Markdown twin. Append .md to any URL on this site — this page, the docs, anything — and you get clean Markdown instead of HTML.
• There is an llms.txt carrying the whole API surface in a form written for language models. Paste it into Claude, ChatGPT or Cursor and the assistant already knows every endpoint and parameter.

If you would rather the agent call the API itself instead of writing code for you, there is a hosted MCP server: see WhatsApp MCP server.

Next

• Full documentation — every endpoint, every parameter
• WhatsApp API pricing — what Meta bills, and what is free
• Official vs unofficial API — which one these endpoints should run on