"Does it integrate with my management system?": the three paths

The short answer: there are three paths, not one

Integrating WhatsApp with my management system is not a single thing, which is why the "yes, it integrates" of demos does not help.

There are three paths, and the one that applies depends on what system you use and who can touch it:

Path	What you need	How much work it is
Ready-made integration	Your system to be supported	Pick it in the panel and enter credentials
Your own API	An endpoint that already exists	Configure one skill per endpoint
An MCP server	An MCP server running	One connection, and it brings all its tools

All three end up in the same place: the conversation can query and do things in your system. The path changes, not the destination.

Path 1: the ready-made integration

The shortest one, and it only exists if your system is supported.

The first one supported today is Dux Software, widely used in Argentina, and the rest get added one at a time. If yours is there, the connection is made from the panel: pick the provider, enter your credentials, look at the preview of what will be imported, and switch it on.

No ticket, no waiting, no technical meeting. That is the whole point of having it ready-made.

Path 2: your own API

If your system publishes an API, that API can become something the conversation knows how to do.

It works like this: each skill points at one endpoint. Checking a balance is one skill, looking up a delivery note is another. It is configured from the panel and can be enabled on as many services as you like.

Almost every management system has this and almost nobody uses it. Dux, for example, documents that over its API "it is possible to create orders, check prices and stock", and clarifies in the same guide that the integration is development external to their system. The door is there; somebody has to walk through it.

Path 3: an MCP server

This is the new path, and the one fewest people understand from the business side.

An MCP server is connected once and hands over all the tools it exposes, in one go. Where with an API you build one skill per endpoint, here you connect the server and everything that server knows how to do shows up.

If your software provider already publishes one, it is the shortest of the three paths after the ready-made integration.

What MCP is, without the jargon

MCP is an open standard for connecting AI applications to external systems.

The official documentation explains it with an analogy that works: it is like a USB-C port for AI applications. Just as USB-C standardises how devices connect, MCP standardises how an AI connects to data and tools.

For your business the translation is simple: the connector gets written once and works for any compatible system. You are not left depending on some provider having built the integration with you specifically.

Importing from an OpenAPI: the shortcut

If your system publishes an OpenAPI specification, there is a fourth move that saves half the work of path 2.

You paste the specification URL, tick the endpoints you want, and the skills come out already configured, one per endpoint. The fields arrive filled in from the specification.

One detail that avoids confusion: the specification declares what type of authentication each endpoint uses, never the secret. You supply the credentials.

Testing it before anybody uses it

The part that decides whether this is usable for real.

Every skill can be run from the panel and you see the real result: the status it returned, how long it took, and what it answered. That happens before any conversation uses it.

It is the difference between "it should work" and "I watched it work". An integration tested only in production is tested with a customer in front of it.

What happens to your credentials

They are stored encrypted and never returned. Not to the panel, not to anybody reading the API.

It sounds like a technical detail and it is a business question: you are handing a third party the key to your management system. That this key cannot be read back from outside is part of what you are buying.

What to know before getting excited

Three limits, said now and not in the third meeting.

• It is a Pro-plan-and-up capability. Custom skills are not on every plan.
• One API, one endpoint per skill. It is not one endpoint that does everything.
• One MCP server per skill. You connect the server, not a list of servers.

Key takeaways

• There are three paths to integrating WhatsApp with your system: a ready-made integration, your API, or an MCP server.
• The ready-made one is the fastest and today covers Dux Software; the rest get added one at a time.
• With your own API you build one skill per endpoint; with MCP you connect once and get all the tools.
• MCP is an open standard, officially described as a USB-C for AI applications.
• If your system publishes an OpenAPI, skills are imported by ticking endpoints.
• Everything gets tested from the panel before a conversation uses it, and credentials are never returned.