.png){kind=icon}
Does Exact Online have an MCP server? What exists now and what is coming
Exact puts its own MCP server on the roadmap, without a date, package or price. What that means, what already works today, and what to look out for once it is here.
Short answer: not yet, but it has been announced. On its roadmap page, Exact lists its own MCP server for Exact Online as "coming soon", with this description: "With MCP, AI agents (such as ChatGPT) can retrieve data in real time, run analyses and start actions within Exact Online." No date, package or price is mentioned. What you can do today is connect your administration to your AI assistant through a connector party, which is what we do with Exact AI Connect. In this article I set out the facts without speculating about what Exact has not said.
What an MCP server for Exact Online does
MCP stands for Model Context Protocol, an open standard that lets AI assistants such as ChatGPT, Claude, Copilot and Gemini talk to software in a fixed way. An MCP server for Exact Online is the piece that sits between your AI assistant and your administration. The assistant asks "which invoices have been open for more than 60 days", the server translates that into a request to Exact Online, and the answer comes back in the conversation. You can read exactly how this standard works in <a href="https://www.iwebdevelopment.com/blog/wat-is-mcp-model-context-protocol">What is MCP?</a>.
The main thing to remember is that MCP is a standard, not a product. Whoever builds the server determines what it can do. Two servers for the same package can differ completely in what they read, what they are allowed to write, and how they handle an error.
What Exact has announced
The roadmap mentions the MCP server alongside three AI agents (Financial, CRM and Excel) and the Exact AI Assistant. For the MCP server, this much is certain: it is planned, it targets AI agents that "retrieve data, run analyses and start actions", and it comes from Exact itself. There is nothing more. No date, no edition it belongs to, no price, no list of what it can do exactly, and nothing about multiple administrations or about writing with or without confirmation.
We mainly read the announcement as confirmation. When the supplier of the package itself chooses this route, "connecting your own AI to your bookkeeping" has become the normal way of working, not an experiment. Other Dutch packages were already ahead of Exact. Jortt put its own MCP server live at the end of August. That is good news for everyone waiting for this, including us.
What you can already do today
Through the Exact Online API, connector parties can already offer an MCP server, and that is happening. There are now a handful of them in the Netherlands, with big differences in what they do. Exact AI Connect is ours. What it does today, in one paragraph: you connect your administration once through your Exact login, choose the administrations the AI may see, and then ask your questions in plain English. Reading covers the entire administration. Writing (creating a contact, a booking, a reconciliation) is possible too, but always with a confirmation step, and you can limit write permissions per role or switch them off entirely. For large analyses across the full dataset there is a separate analysis environment, so you do not get stuck on the page size of the API.
In addition, there are things an MCP server from Exact itself will by definition not do, because they lie outside Exact: the same way of working for Twinfield and AFAS, and the modules we build around it, such as purchase invoices read from your mailbox and booked after approval.
What to look for when it arrives
As soon as Exact's server becomes available, this is the list we would work through ourselves:
- Which edition and which price. Is it in every Exact Online edition or only in the higher ones?
- Reading, writing, or both. And if writing is possible: is there a confirmation step, or does the AI act straight away?
- Multiple administrations. Can you discuss two companies in one conversation, or do you have to connect again for each administration?
- Permissions. Can you limit what the AI may do per user or role, or is it all or nothing?
- Which AI assistants. Exact mentions ChatGPT as an example; does it also work with Claude, Copilot and Gemini?
- Large questions. The Exact API returns a limited number of lines per request; how does the server handle "revenue per customer over three years"?
You can put the same questions to us, by the way, and you should expect the same from any other connection provider.
Can you use both side by side?
Yes. MCP is an open standard, so your AI assistant can have several servers at once, just as you have several apps on your phone. If Exact's server arrives and does something yours does not, you simply add it. What you connect today does not have to be thrown away.
Want to try it yourself? Start with Exact AI Connect for free →
Frequently asked questions
Not yet. Exact has announced one on its roadmap page, as "coming soon", without a date, edition or price. Until then you connect Exact Online to your AI assistant through a connector party that offers an MCP server.
An intermediate layer that lets your AI assistant (ChatGPT, Claude, Copilot, Gemini) talk to your Exact Online administration through the open standard MCP. The assistant asks the question, the server pulls the answer from Exact and returns it in the conversation.
Yes, through Exact AI Connect. You connect with your Exact login, choose which administrations the AI may see and then ask your questions in plain English. Writing is possible too, always with a confirmation step.
Not necessarily. MCP is an open standard, so both can exist side by side. What a connector party adds lies in the things outside Exact: the same way of working for Twinfield and AFAS, modules such as purchase invoices from your mailbox, and an analysis environment for large datasets.
At the edition and the price, at reading or also writing and whether a confirmation step is included, at multiple administrations in one conversation, at rights per user, at which AI assistants are supported and at how large queries are handled.
