← Field notes// field note

AI and Moneybird: connect Moneybird to ChatGPT, read-only

Moneybird already did half the work: there's an official MCP server. Here's when that's all you need, and when a custom read-only connector makes more sense.

October 2026

6 min read

You can connect Moneybird to ChatGPT or Claude today. Moneybird offers an official MCP server with a read-only and a read-write endpoint, plus a REST API with OAuth or personal tokens. For one administration, start with the official read-only connection. Build your own connector when you combine Moneybird with other systems, many administrations or strict logging.

AI and Moneybird: connect Moneybird to ChatGPT, read-only, QuikCode field note

Want to use AI with Moneybird? Moneybird already has an official MCP server, so you can connect Moneybird to ChatGPT or Claude without building anything. It has a read-only and a read-write endpoint, and Moneybird itself recommends starting read-only. For a lot of businesses that's enough. For some it isn't, and that's where the Moneybird API and a custom connector come in.

I haven't built a Moneybird integration. I built the same thing for Nmbrs Accounting, for a bookkeeping office with around 246 active client administrations. That's described in connecting Nmbrs Accounting to ChatGPT and in the Landsman en Teune case. This note is what I learned there, applied to what Moneybird publicly offers.

Option 1: Moneybird's own MCP server

MCP (Model Context Protocol) is the open standard that lets an AI assistant use tools. Moneybird hosts its own MCP server, with two addresses:

You connect it with OAuth: you approve the connection in Moneybird and choose which administration to share. Moneybird lists Claude Desktop, ChatGPT and Cursor as supported, and any other client that speaks remote MCP. Their own example questions are things like "how much is still outstanding?" and "which customers have been paying more slowly lately?".

Moneybird's own advice, which I'd repeat: start read-only, review anything the assistant drafts before confirming it, and revoke the connection if a laptop goes missing. For unattended, scheduled automation they point to the API or their command-line tool instead of MCP.

If you run one Moneybird administration and want to ask questions about it in ChatGPT, stop here and use this. Check the Moneybird help center for the setup steps.

Option 2: the Moneybird API

Under the MCP server sits the Moneybird REST API. If you want to build anything yourself, these are the facts that matter:

  1. One URL pattern per administration. Every call goes to moneybird.com/api/v2/ plus the administration ID, then the resource. JSON in, JSON out.
  2. Two ways to log in. OAuth for an app that other people connect, or a personal API token for your own use. Both use scopes: sales invoices, documents, estimates, bank, time entries and settings. You create tokens in Moneybird under the settings for external and AI integrations.
  3. Scopes are by area, not by read or write. In the API docs I couldn't find a read-only option for tokens. So if you build your own and want read-only, enforce it in the connector: only offer tools that read. That's exactly how I did it for Nmbrs, and it holds whatever the token allows.
  4. There's a rate limit. 150 requests per 5 minutes per IP address, and 50 per 5 minutes for the reports endpoints. Go over it and you get a 429 with a "retry after" header. An AI assistant that chains tools uses calls fast, so caching matters.
  5. Lists come in pages of at most 100 items.
  6. There are sandbox administrations. You can build and test without touching real books, and without a paid subscription. Invoices in a sandbox get a watermark. (Nmbrs Accounting and Accelerate didn't give me that luxury.)
  7. GET responses are cached. Moneybird warns you can't rely on a read straight after a write to confirm a change. Check the status code instead.

There's also a community-built, self-hostable Moneybird MCP server on GitHub. It's not made by Moneybird, so read the code before you give it a token.

When you'd build your own connector

This is the pattern that worked for Nmbrs, and why the official route wasn't the answer there.

Is your bookkeeping data used for AI training?

That depends on the AI account, not on Moneybird. On a business account (ChatGPT Business, a Claude Team or Enterprise plan, or the API), your data is not used to train models by default. On a personal account, check the data settings before you connect your books. In the Nmbrs build, the AI runs in the client's own ChatGPT Business account and the connector makes no AI calls of its own.

What you can connect next to Moneybird

The same read-only pattern works for most systems with an API. Combinations I'd look at first for a Moneybird user:

More builds like this are on my work page.

When this is not the right approach

Moneybird made the first step easy. After that you decide which data, for whom, read or write, and what gets logged. If you want Moneybird to answer questions next to your other systems, let's talk for 30 minutes.

Using another package than Moneybird? Here is what exists for Nmbrs, Twinfield, AFAS and SnelStart, and when a custom connector makes sense: Claude or ChatGPT on your accounting software.

Does Moneybird have an MCP server?

Yes. Moneybird hosts an official MCP server with a read-only and a read-write endpoint. You connect it with OAuth and choose which administration to share. Deleting data isn't supported through it.

Can I connect Moneybird to ChatGPT?

Yes. Moneybird lists ChatGPT, Claude Desktop and Cursor as supported clients for its MCP server. Start with the read-only endpoint, so the assistant can answer questions without changing anything in your books.

Does Moneybird have an API?

Yes, a REST API (version 2) with OAuth or personal API tokens and scopes per area, such as sales invoices, documents, bank and time entries. There are sandbox administrations for testing, and a rate limit of 150 requests per 5 minutes.

Can a Moneybird API token be read-only?

The API scopes are per area, not per read or write, and the API docs don't describe a read-only token. If you build your own connector, enforce read-only in the connector by only offering tools that read. The official MCP server has a separate read-only endpoint.

Is my Moneybird data used to train ChatGPT or Claude?

Not on a business account or the API, where data is not used for training by default. On a personal account, check the data settings before you connect your administration.

When should I build my own Moneybird connector?

When Moneybird is one of several systems you want to ask about in one question, when many people or many administrations are involved, or when you need your own logging and hosting rules. For one administration, the official MCP server is enough.

Got a process that should be software?

Book a call
© 2026 QuikCode · software for better operations