Send your first WhatsApp message in the sandbox

Esta página aún no está disponible en tu idioma.

Before you test, create a project and choose who writes its customer replies:

Wabery AI agent — Wabery hosts and runs the chatbot.
External agent — your backend, n8n, Make, or Zapier receives webhooks and sends replies through Wabery.

Not sure which one fits? Read Getting started: choose who writes the replies first.

Now choose how you want to run the setup:

AI coding agent (recommended) — connect ChatGPT, Claude, Claude Code, Cursor, or Codex and tell it what you want to configure.
Wabery website — use Project settings and Inbox without installing anything.

Both use the same project. Switch whenever you want.

Want to stay in the dashboard? Skip straight to the website tutorial.

Before you start

You need:

a Wabery account and project; new signups create the first project during onboarding
a configured Wabery-hosted agent or external reply backend
a phone with WhatsApp installed
an MCP-capable AI app for the recommended path

You do not need a business number, Meta setup, an API key, or the Wabery CLI. The sandbox is enough.

Create a free project

Path 1: Set up with an AI coding agent (recommended)

1. Connect your agent

Open Access → Connections in Wabery and choose Add connection. The website shows the current setup steps for each supported client:

Remote AI app — ChatGPT or Claude, connected through OAuth
Local agent — Claude Code, Codex, or OpenCode

You can also connect a local MCP client directly to Wabery’s hosted server:

claude mcp add --transport http wabery https://mcp.wabery.com

Run /mcp in Claude Code and sign in to Wabery.

Add this to ~/.codex/config.toml:

[mcp_servers.wabery]
url = "https://mcp.wabery.com"
auth = "oauth"

Then sign in:

codex mcp login wabery

Open Customize → MCP, add a server with the URL below, then select Needs login:

https://mcp.wabery.com

The hosted connection uses OAuth, so there is no API key to copy and no Wabery CLI to install. See the MCP guide for other clients, permissions, and local setup.

2. Give the agent the goal

Paste this prompt into the connected agent:

Use the Wabery MCP server to help me complete the sandbox quickstart.

1. List my Wabery projects and use the project I select.
2. Check whether it uses a Wabery-hosted agent or an external agent and explain what will answer my test message.
3. Check that the selected reply path is ready. Do not change the mode without asking me.
4. Find its sandbox join code and sandbox WhatsApp number.
5. Tell me exactly what to send from my phone to join the sandbox.
6. Wait for me to send a normal WhatsApp message after joining.
7. Find the new conversation and confirm whether the configured agent replied.
8. If it did not, diagnose the setup without publishing anything to Live.

Ask before making changes and do not publish anything to Live.

The agent will inspect the project and give you the join code and sandbox number. Send the join code from your phone, then send a second, normal message such as hello from quickstart to the same WhatsApp chat.

The join code only registers the phone; it does not appear as a conversation message. The second message creates the conversation the agent can reply to.

3. Approve and verify the reply

Within a few seconds, the configured hosted or external agent should reply in the same WhatsApp chat. The MCP agent can inspect the result and help diagnose anything missing.

You can also open Inbox to see the new conversation and its message history. High-risk actions require confirmation, and publishing a hosted agent to Live always remains a website action.

Path 2: Use the Wabery website

Choose this path when you want to test the sandbox without connecting an AI agent.

1. Check who will reply

Open Agent. The page shows the project’s current mode:

Wabery AI agent — finish the guided setup and make sure a Draft is ready for sandbox testing.
External agent — finish the checklist, verify the webhook, and turn on Use my API.

Do not continue until the selected path is ready. The sandbox carries messages; it does not create the reply logic for you.

2. Find the sandbox details

Open Channels, choose the WhatsApp Business sandbox row under Testing, and find:

the shared sandbox WhatsApp number
your project’s join code, formatted like wab-ABC123

Use the WhatsApp button when it is available, or send the join code to the sandbox number manually.

3. Start a conversation

After WhatsApp confirms the join code was sent, send a second, normal message to the same chat:

hello from quickstart

Open Inbox. The new conversation appears after that normal message arrives. If Inbox is still empty, check that you sent the normal message after the join code.

4. Check the automatic reply

Your hosted or external agent should answer within a few seconds. Open the conversation in Inbox to inspect both messages.

If you want to step in, send a manual reply from Inbox. That turns on handoff for the conversation so automation does not talk over you.

Continue building

The quickstart proves your chosen reply path works. Add only what the project needs next:

Goal	Next step
Improve a hosted agent	Add knowledge, automations, Flows, or a hosted function, then test the Draft again
Improve an external agent	Add signature verification, idempotency, media handling, and production error handling
Let an AI coding agent help	Keep the MCP connection and describe the next change
Work directly from a terminal	Install the optional Wabery CLI

Next steps

Connect an AI agent

Use Wabery from ChatGPT, Claude, Cursor, Codex, or another MCP client.

Webhooks

Receive inbound messages and delivery statuses as signed events.

Sending messages

Send text, media, templates, and interactive messages.

WhatsApp chatbot sandbox

Demo a client’s chatbot in the sandbox, then go live on their number.

Authentication

Create server-side API keys when your own application needs one.

Connect WhatsApp

Move from the shared sandbox to a dedicated business number.

Open Wabery