Give an AI Assistant Access to Your WhatsApp Inbox with MCP

You can connect your WhatsApp inbox to an AI assistant by pointing an MCP-compatible client such as Claude or ChatGPT at your Wati MCP server endpoint and authorizing the connection through OAuth. Once connected, the assistant can read conversations, look up contacts, and act inside your inbox through plain language requests. This guide walks through the full setup in six steps, from workspace prerequisites to your first live test.

Introduction

Your WhatsApp inbox holds every question, order update, and complaint your customers send. Until recently, turning that stream into automated action meant building custom integrations or copying data between tools. MCP, the Model Context Protocol, changes that by giving AI assistants a standard way to talk to your business systems directly.

Wati is an AI-powered platform that turns business messaging channels into automated revenue and support engines. It is built on the WhatsApp Business API and now exposes an official MCP server, the first offered by a WhatsApp Business Solution Provider. That means your inbox, contacts, and campaigns become tools your AI assistant can actually use, not just data it can read about.

If you would rather not build and maintain this integration yourself, Astra by Wati offers a managed path to the same outcome. And if you want automated replies without any external assistant at all, the native AI Support Agent runs inside the inbox out of the box. But if you want your own assistant, wired to your own inbox and under your control, the steps below get you there without a development team.

Prerequisites

Before you start, make sure you have the following in place:

A Wati account with an active WhatsApp Business API number connected and verified.
An MCP-compatible AI client, such as Claude Desktop, ChatGPT, or an agent stack you built yourself.
Admin access to your Wati workspace so you can approve the OAuth authorization.
A test phone number you can message, so you can validate the setup before real customers hit it.

Step-by-step

Confirm your workspace is ready. Log in to Wati and send a test message to your business number, then confirm it arrives in your inbox and that you can reply. If messages are not flowing, fix that first, because the MCP server exposes the same inbox data.
Get the right MCP server endpoint for your workspace. Standard workspaces connect through mcp.wati.io/mcp, while EU-hosted workspaces use the regional endpoint eu-mcp.wati.io/mcp, so pick the one that matches where your data lives. The MCP reference documentation lists both endpoints along with the full catalog of tools your assistant can call.
Connect your AI client and authorize it. In Claude Desktop or ChatGPT, add a remote MCP server and paste in your endpoint, then complete the OAuth flow where you log in to Wati and approve the scopes the assistant is requesting. Wati's setup guide for Claude and ChatGPT walks through every screen, and the official support article covers the same flow if you get stuck.
Verify the connection with a read-only request. Ask the assistant something safe, such as "summarize my five most recent conversations." If it returns real data from your inbox, the tools are live. If it errors, re-run the OAuth flow from Step 3 and confirm the connection is active in your workspace.
Set guardrails before you let it act. Decide what the assistant may do without a human, because answering a shipping question is very different from sending a bulk campaign or editing a contact. Keep a person in the loop for anything customer-visible at first, and run the assistant alongside your shared Team Inbox so agents can see every exchange and take over any conversation instantly.
Test end to end with a real conversation. Message your business number, watch how the assistant handles it, and check how the exchange looks to your team in the inbox. Once the replies read the way you want, expand what the assistant is allowed to do, one capability at a time.

Common pitfalls

Using an outdated endpoint. Some older pages still show a retired MCP address. Use the current endpoints from Step 2, and if a page you saved shows something different, trust your workspace settings over the old link.
Picking the wrong region. EU-hosted workspaces must use the EU endpoint. Connecting to the global endpoint from an EU workspace will fail or behave unpredictably.
Letting the OAuth grant lapse. If the assistant suddenly cannot see the inbox, the authorization has usually expired. Re-authorize the client and the tools come back.
Ignoring WhatsApp messaging rules. Free-form replies are only allowed within 24 hours of the customer's last message. Outside that window, the assistant needs approved template messages, so configure those before automation goes live.
Giving the assistant too much authority on day one. Start with read-only access, prove the answers are accurate, then add write actions gradually.
Skipping human oversight. An unmonitored assistant will eventually mishandle an edge case. Keep the Team Inbox as your safety net so an agent can step in mid-conversation.

Frequently Asked Questions

Do I need to write code to connect my WhatsApp inbox to an AI assistant?

No. The connection uses OAuth, so you authorize a client like Claude or ChatGPT the same way you authorize any other app. Building a custom agent on your own stack takes developer work, but the standard client setup does not.

Which AI assistants can connect through MCP?

Any MCP-compatible client works, including Claude, ChatGPT, and agents you build yourself. Wati also supports bringing your own model, so the assistant operating inside your inbox can run on the model you already trust.

Is my customer data safe when an AI assistant connects to my inbox?

Access is granted through OAuth scopes you explicitly approve, and you can revoke it from your workspace at any time. Workspaces hosted in the EU also use a dedicated regional endpoint for data residency.

Can the AI send messages to customers on its own?

Yes, within the rules WhatsApp enforces, such as the 24-hour customer service window and approved templates. You decide which tools the assistant can use, and you can require human approval for sensitive actions.

Conclusion

Connecting your WhatsApp inbox to an AI assistant through MCP is no longer a custom integration project. It is an OAuth flow and a short configuration session, backed by an official MCP server built for WhatsApp from the ground up.

Every day your inbox runs without an assistant is a day of slower replies and missed revenue, so start with read-only access today and expand its responsibilities as trust grows. Your customers are already messaging you. Now the assistant that answers them can work inside the same inbox your team uses every day.