AI to Dinero integration: API, MCP and what pays off

AI to Dinero integration pays off most where a person currently reads and types: purchase receipts that need interpreting, invoice drafts that need creating, customers and payments that need moving between Dinero and the other systems you work in. The booked entry should still be approved by a human. That is not a limit of the technology, it is a choice you should make deliberately.

What is new in 2026 is that there are two ways in. Dinero has a REST API that has been around for years, and on top of it an official MCP server in beta that connects an AI assistant directly to your books. The two solve different problems, and they do not cost the same.

We build this kind of integration for Danish SMBs. The rest of this guide is written from what goes wrong in the first attempts, not from a feature list.

AI to Dinero integration: the three levels

There are three levels, and they cost very different amounts. Most SMBs need level two, but get sold level three.

AI only matters when something needs interpreting. A bank reconciliation does not need a language model. A PDF from a supplier that follows no template does. So does an email where a customer asks for a credit note without mentioning the invoice number, and an account string that has to be guessed from a product description.

The decision to use AI is no longer early, by the way. 59 percent of Danish companies with more than 10 employees used AI in 2026, and among those with 10 to 49 employees it was 55 percent, according to figures from Statistics Denmark. The question is rarely whether, but where in the chain.

What Dinero handles itself before you build anything

The cheapest project is the one you never start. On the higher plans, Dinero reads the date, amount and supplier from a purchase receipt, suggests an account based on your earlier entries, reconciles the bank and puts its suggestions in Tasks, where you approve them. You can also set up automatic bookkeeping rules so that regular suppliers land correctly without anyone touching them.

The next step is the app catalogue. It holds ready-made integrations for payroll, POS, webshops and payments, and they are typically set up in an afternoon. A ready-made app that covers most of your need almost always beats a custom solution that covers everything.

Ask yourself three questions before moving on to the API:

• Is there an app in Dinero's catalogue that covers most of the task?
• Is the rest something a person can handle in under an hour a week?
• Can the process be changed so the need disappears?

If you answer yes to any of them, stop. We say the same to people who call wanting something built that they could simply subscribe to.

Dinero's official MCP: what it can do and where it stops

This is where Dinero stands out from its competitors. Where e-conomic has no official MCP server, Dinero has opened its own in beta. You point your AI assistant at Dinero's MCP address, log in, approve access with your Dinero user, and then Claude or ChatGPT can work in the books. It is a Pro and Total feature, the access requires your approval every time it is set up, and you can remove the connection again.

It is strong at what an assistant is good at: answering how much a particular customer owes you, finding the invoice nobody remembers the number of, preparing a VAT reconciliation, drafting a payment reminder.

There are three things to know before you connect it.

Access follows the user, not the company. If you approve with a user that has access to several Dinero companies, they all come along. If you have your own books and a customer's on the same user, think about that twice.

It is beta, and the responsibility is yours. Dinero takes no responsibility for what an AI types wrongly, and the MCP will not catch it for you. In practice that means you review the entries yourself afterwards, exactly as you would with a new bookkeeping trainee.

An assistant is not an operational flow. MCP works when you are sitting there asking questions. A flow that has to run every night, keep track of what it has already processed, log its errors and raise a flag when something looks wrong is not a conversation. It is an integration, and that is level three. The logic is the same as in our guide to AI to e-conomic integration, but the access routes are completely different.

Two access routes to Dinero's API

This is where it gets technical, and where most Danish articles stop. Dinero has two separate access systems, and you only need one of them.

Personal integration or Visma Connect

Personal integration is the route when you are automating your own books. You go to Integrations in Dinero, open the overview of API keys, choose Personal Integration and request credentials. You exchange those for a token at Dinero's authorisation service and then call the API. You create one key per organisation, and Dinero's own documentation recommends creating a test organisation first and building against that. This route requires an active Pro or Total subscription.

Visma Connect is the route if you are building a product that other companies should be able to connect to their own Dinero. You register the integration in Visma's developer portal, request the right scopes, and Dinero approves it technically before it can be used. It is more work, and it is only relevant if you are reselling the integration.

Dinero also sets a ceiling. Partners with several customers may make 100 calls per minute per organisation, and personal integrations 60 calls per minute, before you get HTTP 429 back. That is stated in Dinero's own notice on rate limits, which also mentions webhooks for entries and invoices. If you hit the ceiling, the answer is almost always that the flow is asking instead of listening.

Four automations we build on Dinero

These are the four that most often pay for themselves. All four have a human as the last step.

Receipt to draft. An incoming invoice lands in a mailbox. A model reads supplier, amount, VAT and date, suggests an account based on how the same supplier was booked the last few times, and puts a draft in Dinero. The bookkeeper approves. The time goes from a few minutes per receipt to one click.

Debtor follow-up with language. Dinero can send payment reminders. What it cannot do is write differently to a customer who always pays on time and one who owes for the third time. A flow fetches the overdue items, groups them per customer and writes a proposal that you approve.

Reporting across systems. Figures from Dinero are rarely enough on their own. Combined with hours from the project system or orders from the webshop, they become a contribution margin per customer. It is the same logic as in automated reporting, just with the accounts as the source.

Two-way data sync. A new customer created in the CRM should exist in Dinero without anyone retyping it, and a paid invoice should be visible in the sales system. It sounds trivial, and it is where the hours disappear in an SMB. We have written more about it in the guide to automating data entry.

Where the limits of AI to Dinero integration lie

Four limits are worth knowing before you decide anything.

Draft, not booking. Let an agent create drafts and suggestions, and let a human press approve. That is not because the model is bad. It is because a mistake in a draft is one click to fix, while a mistake in a booked entry is a correction in the accounts that has to be explained afterwards.

The subscription is a prerequisite. Both API keys and MCP sit on Pro and Total. If you are on a lower plan, the first bill is not development, it is a change of subscription. Worth knowing before anyone promises you a solution in two days.

Do not ask, listen. A flow that checks Dinero every minute to see whether anything has happened burns calls without achieving anything. With several companies on the same key you approach the ceiling. Webhooks solve it, because Dinero then tells you itself.

Personal data. Accounting data contains names, addresses and payment patterns. Sending them to a language model is a transfer you must be able to explain. We therefore work with data minimisation as the starting point: only the fields the task requires, and a model in the EU where possible.

What does a Dinero integration cost?

A well-defined flow, meaning one of the four above, starts at 10.000 kr. as a one-time fee with us. If we also look after it on an ongoing basis and keep building, it is from 5.000 kr./month. You own the code and the documentation, so you can take it with you if you want something else.

The calculation you should make is not price against price. It is price against hours. A team that spends 15 hours a week typing and checking spends more than a half position on it. A full-time hire costs 25.000+ kr./month and does it every month. An integration is built once.

Key numbers

15 hours

we typically find per client each week

20.000 kr./month

saved at Markeity on systems built once

53

automations in production

Half of the advisory cost may be covered if the project falls under SMV:Digital, which pays 50 percent of the advisory cost and up to 150.000 kr. We have set out the pools and the requirements in the guide to SMV:Digital grants for AI, and the figures behind a flow are in the guide to what AI automation costs. If several systems need to talk to each other at once, that falls under system integration.

How we test before anything goes live

We do not build against your real bookkeeping from day one. First against a test organisation, then against a copy of your own receipts, then in parallel running where the flow suggests and a human compares with what would otherwise have happened. Only after that does it take over. From audit to operation is typically about 3 weeks, and a large part of the time goes on finding out how the bookkeeping actually works at your company, as opposed to how it ought to work.

FAQ

Can AI book automatically in Dinero?

Technically a flow can write to Dinero, and the MCP can work in the books. We do not recommend it as a starting point. Let AI create the draft and let a human approve. Once the flow has run for a few months and the error rate is known, you can move the most predictable receipt types over to running on their own.

Do Dinero's API and MCP require a Pro subscription?

Yes. Both API keys for a personal integration and Dinero's MCP sit on Pro and Total. It is the first thing you should check before planning anything, because it shifts the budget before a line of code is written.

What is the difference between Dinero's MCP and an integration built on the API?

MCP is an assistant you ask. An integration is a flow that runs. MCP requires someone to sit there and read the answer, while an integration has logic, error handling and a log so it can run unattended. The two can happily live side by side: the assistant for investigating, the flow for what happens every day.

Can I use n8n, Make or Zapier for Dinero?

Yes, for simple flows. All three can talk to Dinero through HTTP calls, and for a chain of three steps that is faster than building it yourself. If the flow branches, or a model has to interpret something, the complexity comes back somewhere else. We have compared the three in n8n vs Make vs Zapier.

How many calls may my Dinero integration make?

Dinero states 100 calls per minute per organisation for partner integrations and 60 calls per minute for personal integrations, after which you get a 429 back. Use webhooks instead of asking at fixed intervals, and the ceiling is rarely a problem.

AI to Dinero integration is not a question of how smart the model is. It is a question of which links you dare to let go of, and which you want to keep approving yourself.