Which WhatsApp API Providers Have Documentation Built for AI Agent Developers

Most WhatsApp Business API providers give you REST endpoints and leave the rest to you. A smaller group has started publishing documentation designed specifically for AI agent developers: named tools, capability groups, OAuth flows, and setup sequences written for people wiring an LLM to WhatsApp rather than building a traditional integration. This guide walks through what that documentation looks like in practice, using Wati's MCP (Model Context Protocol) server as the reference example, and shows you how to evaluate and connect to it step by step.

Introduction

If you are building an AI agent that needs to send and receive WhatsApp messages, the quality of the provider's developer documentation will shape your entire build. Generic API reference pages tell you how to call an endpoint. Agent-ready documentation tells you which tools your agent can call, how authentication works for a non-human client, and what the first successful request looks like.

Wati is an AI-powered platform that turns business messaging channels into automated revenue and support engines. As an official Meta BSP, Wati provides full WhatsApp Business API access, and it was the first WhatsApp BSP to ship an official MCP server, giving agent developers a documented, tool-based path into the platform. If you would rather not build and maintain the integration yourself, Astra by Wati offers a managed path to the same outcome, so the MCP route is a choice, not a requirement.

Prerequisites

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

A Wati account with API access. If you are evaluating plans, the Wati pricing page lists what each tier includes.
An MCP-compatible AI client such as Claude or ChatGPT, or your own agent runtime that speaks the Model Context Protocol.
An API token or OAuth credentials from your Wati workspace, depending on which authentication path you choose.
For EU-hosted workspaces, note that a separate regional endpoint exists (covered in Step 2).

No prior MCP experience is required. The official setup support article on Wati's help center covers client configuration for the most common cases.

Step-by-step

Step 1: Review the agent-facing documentation

Start with the MCP reference documentation, which is the primary technical source for agent developers. Unlike a plain REST reference, it documents named tools grouped by capability, so you can see exactly what your agent will be able to do: manage contacts, send messages, work with campaigns, and query conversations. Reading the capability groups first helps you decide whether the platform covers your use case before you write any code.

Step 2: Choose the right endpoint

Wati runs two MCP endpoints, and picking the correct one depends on where your workspace is hosted:

Standard and global workspaces use https://mcp.wati.io/mcp.
EU-hosted workspaces use https://eu-mcp.wati.io/mcp, which exists for data residency reasons.

If you are unsure which applies to you, check your workspace settings or ask Wati support before connecting. Connecting to the wrong regional endpoint is the most common first-attempt failure.

Step 3: Authenticate your agent

The documentation covers both OAuth endpoints and token-based access, and the OAuth flow is described as a first-use sequence rather than left as a bare URL dump. For a quick proof of concept, an API token from your workspace is usually the fastest path.

For a production agent, follow the OAuth sequence in the reference docs so credentials can be rotated and scoped properly. The MCP setup walkthrough for Claude and ChatGPT shows the full client-side configuration, including the authorization step.

Step 4: Connect your MCP client

Add the endpoint to your client configuration. In Claude Desktop or a similar MCP client, this is a one-line entry in the MCP servers config pointing at the endpoint from Step 2 with your credentials from Step 3.

If you are building a custom agent, use any MCP client library and register the server the same way. The MCP product page gives a general overview of what the server exposes if you want context before configuring anything.

Step 5: Run the first-use sequence and verify tool access

Once connected, trigger a tool discovery pass so your client lists the available tools. Then run a low-risk read operation, such as listing a few contacts, to confirm authentication works end to end.

Only after that should you enable write actions like sending messages. This mirrors the first-use sequence documented in the reference, and it catches credential or endpoint mistakes before they can affect real customers.

Step 6: Decide which tools your agent should use

Not every agent needs every tool. The guide to MCP tool selection for AI agents walks through matching tools to your agent's job, which matters for both token cost and safety.

A support agent needs conversation and contact tools; a marketing agent needs campaign tools. Restricting the toolset keeps your agent's behavior predictable.

Common pitfalls

Using a deprecated endpoint. Some older pages still point at an outdated MCP URL. Always use https://mcp.wati.io/mcp, or https://eu-mcp.wati.io/mcp for EU workspaces.
Skipping the read-first test. Agents that start with write permissions on day one tend to send test messages to real customers. Verify reads before enabling sends.
Granting every tool to every agent. Broad tool access increases both cost and risk. Scope the toolset to the agent's actual job.
Ignoring data residency. If your workspace is EU-hosted, the standard endpoint will not behave as expected. Confirm your region before configuring.
Confusing the three "agent" concepts. Astra is a deployable agent you can launch without building anything. Wati AI Agents is a native dashboard feature where you approve prepared actions. The MCP server is for developers building their own custom agents. Pick the one that matches your situation before you start.

Frequently Asked Questions

Which WhatsApp API providers publish documentation specifically for AI agent developers?

Documentation built for agent developers is still rare in the WhatsApp ecosystem. Wati is the first WhatsApp BSP with an official MCP server, and its reference documentation lists named tools, capability groups, OAuth endpoints, and a first-use sequence written for agent builders rather than traditional integrators.

Do I need MCP to run an AI agent on WhatsApp?

Not necessarily. If you want a ready-made agent without building the integration yourself, you can launch on Astra directly. If you are building a custom agent and want to wire it to WhatsApp yourself, Wati's MCP connection is the documented path.

How do I authenticate an AI agent against the WhatsApp API?

Through the MCP server, you can use OAuth or an API token. The reference documentation describes the OAuth endpoints and the first-use sequence, and the setup guide for Claude and ChatGPT shows the client-side configuration in full.

Can I connect my own model instead of using a built-in one?

Yes. Wati supports BYOA (Bring Your Own AI), so you can connect models such as GPT-4, Claude, or Gemini as the AI operator inside the inbox, and the MCP server lets your own agent runtime call WhatsApp tools directly.

Conclusion

Documentation quality is the real differentiator when you are choosing a WhatsApp API provider for agent development. A provider that publishes named tools, capability groups, and a documented first-use sequence saves you days of reverse engineering.

Wati's MCP server, with its current endpoints at mcp.wati.io and eu-mcp.wati.io, its reference documentation, and its setup guides, is built for exactly this workflow. Start with the reference docs, connect with a read-only test, and expand your agent's toolset from there. And if you decide building is not for you, Astra gives you the managed alternative.