For AI agents and automation

WhatsApp for AI agents: REST API, n8n node and MCP

An AI agent needs three things to use WhatsApp: tools to send, a way to receive and safety rules. See what ZapZap offers today, what does not exist yet and how to connect an agent to your instance.

Create account See the API reference

How does an AI agent use WhatsApp?

An agent sends messages by calling ZapZap's REST API as a tool (for example send/text and send/media) and receives replies through a webhook. Authentication is two headers, x-api-key and x-api-secret, and the OpenAPI file describes every endpoint for the agent to read.

Today this works through three paths: the REST API with OpenAPI, the community node n8n-nodes-zapzap for n8n flows and the AI agent built into each instance. ZapZap's MCP server has not been published yet; this page says so clearly in the section below.

Ways to connect an agent to WhatsApp

The table shows what exists today (checked 10/27/2026) and what does not.

Path	Status	What for
REST API and OpenAPI	Available: 150+ endpoints, OpenAPI file in pt, en and es	Any agent or framework that calls HTTP as a tool
n8n-nodes-zapzap node	Available on npm (community node, version 0.1.3), not verified by n8n	Agent flows in n8n, on a self-hosted install
Signed webhook	Available: messages and messages_update events, retries and HMAC signature	Delivering the customer's reply to the agent
Built-in AI agent per instance	Available in the dashboard and the API (agent-edit)	Support without assembling the agent elsewhere
MCP server	Not published yet	When it exists, it would connect MCP clients straight to the tools
LangChain and CrewAI packages	Not published yet	Today, use the API as an HTTP tool

First call, to test the tool

Before putting the API in an agent's hands, test sending with cURL. Create the account, connect a number by QR code and copy the instance ID, the key and the secret.

curl -X POST https://api.zapzapapi.com/api/v1/SEU_INSTANCE_ID/send/text \
  -H "Content-Type: application/json" \
  -H "x-api-key: SUA_API_KEY" \
  -H "x-api-secret: SEU_API_SECRET" \
  -d '{"number":"5511999999999","text":"Olá! Aqui é o assistente da clínica."}'

Give the agent only the tools it needs, with credentials in environment variables and never in the prompt.

Tools you can give the agent

Each row is a real endpoint from the reference. Start with a few.

Tool	Endpoint	Note
Send text	POST /api/v1/{instanceId}/send/text	Fields number and text
Send media	POST /api/v1/{instanceId}/send/media	Fields number, type, file (URL), text, docName
Check numbers	POST /api/v1/{instanceId}/chat/check	Field numbers; returns whether each is on WhatsApp
Get instance status	GET /api/v1/{instanceId}/instance/status	So the agent knows whether it can send
Set the webhook	PUT /api/v1/instances/{id}/webhook	Defines the URL that receives messages

What about MCP?

The Model Context Protocol (MCP) is a standard for exposing tools to AI clients. ZapZap has not published an MCP server: there is no package and no install command, and this page does not make one up.

In the meantime, ZapZap's OpenAPI file has an identifier per operation, which lets you generate an MCP server from it. The post WhatsApp MCP server for Claude and Cursor shows the design. When there is an official server, it will be announced here.

Limits and best practices for agents

Send only to people who agreed to receive (opt-in). An agent that writes to people who did not ask raises the risk of reports and bans.
Set a pace: the agent should respect intervals between messages and per-contact limits. There is no number of messages a day that is guaranteed safe.
Keep a human in the loop: handoff rules to an agent, confirmation before irreversible actions and a record of every send.
Keep credentials out of the prompt and give the agent only the tools it needs. Destructive operations, such as deleting an instance, should not be available to it.
Remember that ZapZap is an unofficial API: it goes against WhatsApp's terms and the number can be limited or banned. The anti-ban feature reduces the risk, it does not remove it.

Frequently asked questions

Does ZapZap have an MCP server?

Not yet. We have not published an MCP server or an install package. Today the agent uses the REST API as a tool, the n8n-nodes-zapzap node or the instance's own AI agent.

Can I use the API with any agent framework?

Yes. If the framework calls HTTP as a tool, it calls ZapZap's API with the x-api-key and x-api-secret headers. The OpenAPI file describes each endpoint's fields.

Is the n8n node verified?

It is a community node (n8n-nodes-zapzap, version 0.1.3 on 10/27/2026), not yet verified by n8n. Unverified nodes only work on self-hosted n8n, according to n8n's documentation.

How does the agent receive the customer's reply?

By webhook: set your application's URL and turn on the messages event. Validate the signature and ignore messages with fromMe true.

Does using an agent change the ban risk?

It neither removes nor creates a new risk, but an agent can send too fast or to people who did not ask. Set a pace, opt-in and limits, and remember that no unofficial API guarantees the number will not be banned.

How much does it cost?

R$15 per active number per month, prepaid, no per-message fee. The built-in agent's AI usage is debited from the balance. The cost of the model behind your own agent is with your provider.