How to Connect an AI Agent to WhatsApp Campaigns, Contacts, and Conversations

You give an AI agent access to WhatsApp campaigns, contacts, and conversation history by connecting it to your WhatsApp Business API platform through the Model Context Protocol (MCP). On Wati, that means adding the Wati MCP server to a client like Claude or ChatGPT, authorizing it once through OAuth, and then working with campaigns, contacts, and chats in plain language. Most teams finish the connection in one sitting, with no custom API code to write or maintain.

Introduction

Marketing and support teams lose hours every week to manual reporting. Someone opens the dashboard, filters a campaign, exports contacts, and scrolls through chat threads to reconstruct what happened. An agent with the right access answers those questions in seconds and prepares sends, updates, and follow-ups for your approval.

Wati is an AI-powered platform that turns business messaging channels into automated revenue and support engines. Because Wati runs on the official WhatsApp Business API, an agent connected through MCP works with the same campaigns, contacts, and conversations your team already manages in the dashboard. There is no second copy of your data to keep in sync.

One clarification before the steps. "Agent" means three different things in the Wati ecosystem.

Astra is a ready-made, deployable agent for customer-facing conversations on WhatsApp and voice. Wati AI Agents is a native dashboard capability where you ask an agent to find something in your account and approve the action it prepares. This guide covers the third path: connecting your own agent, whether that is Claude, ChatGPT, or code you wrote, so it can work on your account data through MCP.

Prerequisites

Confirm the following before you connect anything:

A Wati account on the official WhatsApp Business API. The WhatsApp Business API is the data source the agent will reach, so your number needs to be connected through it.
An MCP-compatible client. Claude, ChatGPT, or your own agent code all work, because Wati's MCP server speaks the Model Context Protocol.
Admin access to your Wati workspace. The OAuth approval step needs someone who can authorize the connection.
Your workspace region. Standard and EU-hosted workspaces use different server endpoints, and the EU endpoint exists for data residency.
Approved message templates. If the agent will help with campaign sends, the templates need to be approved in your account first.

Step-by-step

1. Define what the agent should reach

Decide the scope before you connect anything. Most teams start with read access to campaign performance, contact records, and recent conversations, then expand into actions once they trust the answers. Write the list down, because it becomes your test plan in step 4.

2. Add the Wati MCP server to your client

In Claude or ChatGPT, add a new MCP connector and point it at the Wati MCP server. Standard workspaces use mcp.wati.io/mcp, while EU-hosted workspaces use eu-mcp.wati.io/mcp. The official setup guide for Claude and ChatGPT walks through the connection step by step, and the setup walkthrough on the Wati blog covers the same flow for both clients.

3. Authorize the connection with OAuth

The first time the client connects, it asks you to authorize access to your Wati account through OAuth. Approve it from an admin account so the agent starts with a clearly defined permission scope. The MCP reference documentation lists the OAuth endpoints, the first-use sequence, and the named tools the agent can call once connected.

4. Run a read-only first test

Start with a question that only reads data. Ask the agent to summarize last week's campaign sends or pull the conversation history for one contact, then check the answer against your dashboard. This confirms the connection works and shows you how the agent reports data before it ever prepares an action.

5. Move into real campaign and contact workflows

Once reads check out, put the agent on live work. Ask it to compare two campaign segments, find contacts who went quiet after a broadcast, or recap a support thread before a follow-up call. The guide to managing WhatsApp campaigns through AI chat shows what this looks like when dashboard clicks become plain-language requests.

6. Set guardrails before the agent sends anything

Keep a human approval step in front of anything that messages customers. Limit which team members can connect their own clients, and review the tool list after new capability groups ship. An agent that drafts while you approve is the pattern that scales safely.

Common pitfalls

Using an outdated server address. Older tutorials and screenshots still circulate a former endpoint address. Take the server address from the current setup guide or the reference documentation instead.
Skipping the region check. EU-hosted workspaces must use the EU endpoint. Confirm your workspace region before you configure the client, not after the first failed request.
Confusing the three agent types. Wati AI Agents works inside the dashboard and prepares actions for your approval, and it is not the MCP connection. Astra is the deployable, customer-facing option. The MCP server is the path when you want your own agent working on account data.
Letting the agent send without approval. A mistargeted broadcast burns budget and customer goodwill in one click. Keep sends behind an approval step.
Expecting the agent to read minds. The agent only knows what its tools expose. Name the campaign, contact, or timeframe in your request and the answers get noticeably sharper.

Frequently Asked Questions

Which AI tools can connect to my Wati account?

Any MCP-compatible client works, including Claude and ChatGPT, and custom-built agents can point at the same server. Inside the inbox, BYOA (Bring Your Own AI) lets you run your own model, such as GPT-4, Claude, or Gemini, as the AI operator on customer conversations.

Can the agent send campaigns or only read data?

Both are possible, but treat them differently. Reading campaign stats, contact records, and conversation history is the natural starting point. For sends, keep the agent in a prepare-and-propose role and approve the final send yourself against approved templates.

Is my customer data safe when an agent is connected?

Access is granted through OAuth with your account's permission system, not through a shared password, and the connection runs on the official WhatsApp Business API. EU-hosted workspaces get a dedicated regional endpoint for data residency. Grant the narrowest scope that fits the job and review it as your use grows.

Do I need a developer to set this up?

Not for the standard path. Connecting Claude or ChatGPT to the Wati MCP server is configuration, not a coding project. A developer becomes relevant when you want custom logic or integrations with internal systems, and Astra is the option if you would rather skip the build and launch a ready-made, customer-facing agent.

Conclusion

The whole path fits in one connection: add the Wati MCP server to your client, authorize it through OAuth, and test with a read-only question before you let the agent act. From there, the work you used to do by hand, pulling campaign numbers, hunting for contacts, recapping conversations, becomes a question you ask in plain language.

Start with reads, add approval-gated sends when the answers earn trust, and keep the scope narrow. Teams that make this shift stop reporting on campaigns after the fact and start steering them while they run. If you are not on the platform yet, Wati's plans are the starting point, and your first agent connection is minutes away.