// consulting · mcp servers

Your systems become tools.

The ERP, the database and the APIs you already have, exposed once in a format any AI assistant knows how to use. With the permissions you decide.

The Model Context Protocol is an open standard, published by Anthropic in 2024, defining how a model talks to external software. An MCP server is a program that exposes data and actions in that format.

The point is that you write it once. Before MCP, connecting an ERP to an assistant meant one integration per assistant. With an MCP server you define the tools once and any compatible client can use them: Claude, ChatGPT, an agent of your own, your developers' IDE.

For a company, that changes the arithmetic. The work of exposing your data is no longer sunk into whichever vendor you picked: it stays, and it holds for the model that comes out next year too.

What we expose

04
01

The ERP

Orders, customers, items, stock. Read-only to begin with, writes once the logs say they can be trusted.

02

Databases and warehouses

Queries against company data without going through an export. Through dedicated views, because a model should not be looking at raw tables.

03

The APIs you already have

If an integration exists, it becomes an MCP tool without being rewritten: the server translates between your contract and the standard.

04

Permissions, for real

Every tool has a scope and an identity. An MCP server that exposes everything to everyone is an incident waiting to happen, and it is not what we hand over.

Questions we get first

04

What is an MCP server, in one line?

A program that exposes data and actions from one of your systems in a standard format, so an AI assistant can use them as tools without an integration built specifically for it.

How is this different from the APIs we already have?

APIs are written for developers: a model cannot tell on its own which ones exist or when to call them. An MCP server describes them in a way the model can discover for itself, with the types, the limits and the permissions attached.

What is it actually good for inside a company?

Asking in plain language for things that today need a report: how the month went for that customer, which orders are stuck, what is missing from the warehouse. And having an agent ask, not only a person.

Is it risky to give a model access to our systems?

It is if you give it access to everything. So you start read-only, on a narrow scope, with every call logged. Writes get enabled afterwards, one tool at a time.

Tell us the process. We will tell you if it makes sense.

A free 30-minute call, with no commercial follow-up unless you want it.

Book a call