MCP server development
An MCP server gives an AI assistant access to your own systems, like your ERP, CRM or web shop. We build that server on top of your existing API or database. After that, the AI can only do what you allow. An MCP integration costs €5,000 to €15,000 excl. VAT with us.
Want AI to work with your own data? Then it has to be able to reach it somewhere. An MCP server sorts that out properly. First we look together at whether it pays off for you. Would a simple export do? Then we'll just tell you.
What is an MCP server?
MCP stands for Model Context Protocol. It's an open standard for connecting AI applications to other systems. Claude and ChatGPT support it, for example. An MCP server is the piece of software on your side. It offers the AI a few actions, which MCP calls tools. "Look up a customer" can be one, or "show open orders". It can also hand over data as context. The AI sees which tools exist and picks one when it fits the question. Anything the server doesn't offer, the AI can't do. How that works in practice is in MCP: how to connect AI to the software you already use.
Who is an MCP server for?
For businesses with a system that holds their most important data. An ERP, a CRM, a web shop or a database of their own, for example. Your team asks it questions every day and clicks through screen after screen to get answers. Or you copy data into an AI assistant and back again. That's where an MCP server helps.
When don't you need one? When your package already offers an MCP server that does what you want. Or when you only need an overview now and then. An export is often enough for that.
What we build
We build an MCP server on top of your existing API or database. That server only gets access to what your questions need. The AI gets a handful of tools that do exactly what was agreed. The rest of your database stays closed.
We start with tools that only read. Little can go wrong there, and you quickly see whether the AI fetches the right data. Should the AI change something too, like preparing an order? Then a person confirms it first. The official MCP specification recommends the same: there should always be someone who can stop an action.
The server works with the user's own permissions. If someone only sees their own customers in your system, they only see their own customers through the AI too. We log every call, so you can trace what happened. How to work AI further into your own software is on AI in your app or web app.
How a project runs
- The question. Together we pick a few questions your team asks often. That's where it starts.
- Your system. We look at your API or database and decide which tools are needed. And which data is better left out.
- Building. We build the server, starting with read tools.
- Testing. A small group of colleagues works with it. We check whether the answers are right and where the AI picks the wrong tool.
- Expanding. Once reading runs well, tools that change something can be added. Always with a person confirming.
What you need
An API on your system, or access to the database behind it. No API yet? Then that has to come first, and it counts towards the price. You also need a clear use case. "Something with AI" is too vague to build. "Which orders are open for customer X?" is just right. And an AI application that supports MCP, for the people who'll work with it.
What does an MCP server cost?
An MCP integration costs €5,000 to €15,000 excl. VAT with us. Where you land in that range depends on your system and on how much the AI is allowed to do.
- Is there an API already? A clean, documented API saves a lot of work. If one has to be built first, you'll be higher up.
- How many systems? One system is simpler than a server that combines data from your ERP and your web shop.
- Read only, or write too? Tools that change something need more checks and a person confirming.
- Security requirements. Strict requirements for login, permissions and logging take extra time.
How that price is built up is in what an MCP integration costs. Want to look more broadly at what AI can do for your business first? Start at AI solutions.
Can an MCP server work with my ERP or web shop?
If your system has an API, or we can reach the database, usually yes. The MCP server then talks to that API and offers the AI a few tools. Does your system have neither? Then an API has to come first. We'll discuss that with you up front.
Which AI applications work with an MCP server?
Applications that support MCP. Claude and ChatGPT do, for example, and so do development tools like Visual Studio Code and Cursor. How you add your own MCP server differs per application. We work that out together for the tool your team uses.
Will the AI see more of my system than it should?
No. The server only gets access to what your questions need. It works with the user's own permissions, and we only send the fields the AI really needs. Sensitive data stays out. Every call is logged.
What does it cost to have an MCP server built?
€5,000 to €15,000 excl. VAT. You'll be at the lower end if there's already an API, it's one system and the AI only reads. You'll be higher up if an API has to be built first, there are more systems or the AI may also change data. More in what an MCP integration costs.
Which system do you want to connect to AI?
Tell Adem which system you use and what the AI should do with it. Then we look together at whether an MCP server fits. Would an export do? Then you'll hear that too.