How to integrate with DATEV Rechnungswesen: One connection, reading and writing

DATEV Rechnungswesen is the application German tax advisors and companies with in-house bookkeeping use to record bookings. It is the end of the chain rather than a stop on it, and it is the one DATEV product you can query a ledger from, which is why reporting and reconciliation products end up here rather than upstream.

It is also not a cloud product you sign up for. Our documentation is direct about it: “It does not belong to DATEV’s suite of online applications, meaning that it is essentially an on-prem application that can be hosted by a provider or on your own server.” Your customer runs it, or someone runs it for them.

That has an immediate consequence for what can reach it. Exactly one service transfers data into Rechnungswesen directly. Everything else, documents and booking proposals among them, travels through DATEV Unternehmen Online first, where the tax advisor works before anything is booked.

The commercial and coverage view of this system is on the DATEV Rechnungswesen API page, and thevendor level, meaning the whole list of DATEV data services and the interface partner status, is on the DATEV overview.

Most systems in this catalogue have one target system, a few have two. Rechnungswesen has three, and the third one is the reason writing and reading come apart:

• datev-rewe is the short-token route, and its job is to list the companies a login can reach.
• datev-rewe-longtoken is what you work through once a company has been chosen, in both directions.

What a coverage matrix cannot show you is which of the two you end up holding. The matrix tells you that journal entries and a trial balance are readable. It does not tell you that both directions answer on the same account key, so there is no branch in your client between a write path and a read path.

Because every export call needs a company, our documentation recommends the interactive flow here rather than the manual route: the company is selected during authentication and reused for the calls that follow, so nothing downstream has to carry it.

One more detail from DATEV’s side belongs in your data model rather than in your client code.

DATEV’s dataset identifier is unique per API rather than globally, so the same value on a Rechnungswesen API is not necessarily the same entity as on a payroll API. Keeping your own mapping between your entities and DATEV’s is therefore not optional, and DATEV asks you to explain it during the approval meetings.

One inconsistency in the documentation is worth knowing before it costs an hour: in the manual token section the third step is headed “Authenticate to DATEV Unternehmen Online” while the instruction below it names datev-rewe-longtoken. The instruction is the correct half.

One connection reaching both directions does not mean DATEV has one service behind it. Writing and reading are two different DATEV data services, ordered separately by the tax advisor, and each runs on its own Online API. What changed is that both answer on the same authentication rather than on two.

DATEV offers two refresh token variants. The short one lives eleven hours and reaches every dataset the identity may see. The long one lives two years, extends itself on every refresh, and reaches exactly one dataset. Getting the long one requires the offline_access scope, and it is the token behind the connection above.

The part worth knowing before you plan a reporting feature is that the read side depends on the client credentials rather than on anything in your code. The app behind those credentials has to include the export service, otherwise the connection authenticates and the ledger stays empty. Maesn issues credentials with the full set of services for new integrations, which is why this is a sentence in an article rather than a task in your backlog.

The one date to plan for is two years out. DATEV requires that your connection screen shows the calculated expiry date, in at least TT.MM.JJJJ HH:MM form, along with a status and the full name of the person who issued the token. Build a renewal prompt on top of that date while you are building the connection: two years is far enough away to feel like it never happens, and the failure is quiet when it does.

Ten objects are available on this system, and they sort into three groups rather than two. Seven are read-only, two are write-only, and exactly one travels in both directions: the booking itself.

That single overlap is the most useful fact about this integration, and it is also the most easily lost one. It means a posted booking can be verified against the ledger rather than against your own record, which is the difference between a reconciliation feature and a hopeful one.

It also means the round trip crosses the two DATEV data services from the previous section: you write through the Buchungsdatenservice and read back through the Datenservice Export Rechnungswesen, on one account key and one expiry date.

Files are the second writable object, and they are rarer than they look: creating a file is available on four of the 30+ supported systems, and both DATEV products are among them. In DATEV’s own terms this is the document alongside the booking, which is why the write side of this system speaks of a Belegsatz, a document record, and a Beleglink, the link from a booking to its document.

Those two German terms are DATEV’s own, and the second one has a limit worth knowing before you design around it. The link survives the round trip but the document does not: GET /journalEntries returns the id of the linked document in documentId when there is one, so you can tell that a booking has a receipt and which one it is. What you cannot do is pull the file itself back out. Plan on keeping your own copy of anything you need to show a user.

Eight objects are readable through the export system, and two of them are unusual enough to change what a product can offer.

The first is the chart of accounts. It is the read that turns account mapping into a configuration screen instead of a spreadsheet exchange with the customer’s accountant. If you offer any kind of posting feature, this is what lets you show your customer their own accounts instead of asking them to type numbers.

The second is the trial balance, the Summen- und Saldenliste or SuSa in DATEV’s own vocabulary. Measured across all 30+ supported systems, two of them return a trial balance at all.

And there is a second condition that narrows it further: the report endpoint requires a fiscal year start date, and of those two systems only this one also exposes fiscal years. So one system in the catalogue returns the report and hands out its own precondition, and it is this one.

The write row carries a detail that explains why “DATEV format” is a subject in this ecosystem rather than a field. For the Buchungsdatenservice DATEV publishes two separate sets of requirements: one for the data service and one for the DATEV file format itself.

Meeting the first does not mean you have met the second. Through Maesn you post one normalised booking and the translation into that format happens on our side, which is most of the reason this integration is worth outsourcing at all.

Two smaller reads round out the list. Payment terms are available, and they are useful without being decisive. Open items are in beta rather than generally available, so if your product depends on them, treat that as a conversation rather than an assumption.

The reads are filtered and paged the same way as every other system on the platform, which matters more here than on a system you only write to: a ledger is worth paging through.

How that is normalised across systems that each invented their own parameters is the pagination and filtering subject, and a reporting product built on these reads is the shape described on financial analysis and forecasting.

Contacts are the second write-only object, and they behave in a way that has no equivalent elsewhere in this catalogue. You can create a contact. You can update it in place. You cannot read it back.

It is tempting to assume a second route exists, because on most systems it does: where contacts are unavailable you usually reach the same record through customers or suppliers. Not here. Both are unavailable on this system, which makes contacts the object to map a business partner onto and simultaneously the object you cannot verify.

One thing not to read into it. Our own documentation describes creating “journal entries, suppliers and customers” directly in this system, and the object states say otherwise: suppliers and customers are unavailable, contacts is the writable object.

Where the two disagree, the coverage data is what the platform actually enables, and that is what holds here. Which object a business partner maps onto across systems that all name it differently is part of the common data model.

No object on this system has delete enabled, and that is less of a limitation than it sounds in an accounting context. Corrections and cancellations are expected to be new documents rather than removals, so a design that posts a reversal is both more correct and closer to how an accountant will read the result.

Three of DATEV’s requirements shape the design of a read-heavy integration, and none of them is an API detail. They are conditions of the production release, and they are checked.

Change detection cannot be a loop. Counted separately, because the three states of that column mean three different things: no object on this system carries a native event, eight combinations are marked on request and every one of them is a webhook, and the remaining 29 are unavailable.

On the other side DATEV lists round-the-clock polling as a DONT and asks that calls follow something the customer just requested, manually or by a rule. So the shape that is left is a read triggered by an action or by a defensible schedule. The interval is yours; “always” is not one of the options. What a system without events changes about a sync is the unified webhooks subject.

A technical log is mandatory. DATEV requires an HTTP log of requests and responses, kept for at least 14 days, from which the chronological order of calls can be reconstructed.

The minimum is specific: timestamps, the verb and the full URL including query parameters, the headers without the authorization header, and on the response side at least the X-Global-Transaction-ID and V-Cap-Request-ID headers. It does not have to be visible to your customers. It does have to exist when you open a ticket.

The error rate is a number with a price. After production release DATEV expects fewer than ten percent of requests to return 4xx or 5xx, monitors it, and exceeding it triggers a chargeable re-inspection. On a read-heavy integration against a connection that expires twice a year, the obvious way to breach that is a scheduled read that keeps running after the authentication lapsed.

Four things stop being yours:

• Both connections and both lifetimes. Two target systems, two consents and two expiry dates are held per tenant, and your call sites see one account key shape.
• The booking format. You post the common data model; the CSV in the DATEV format, the document link and the file-format requirements behind it are handled here.
• The route selection. Whether a payload can go directly into Rechnungswesen or has to travel through Unternehmen Online is a parameter rather than a second project.
• Access to DATEV without the partner process first. You build on the Maesn client, so nothing has to be applied for, paid or waited on before your first call. If you later want to be an official DATEV Interface Partner in your own name, we run that process with you and you move onto your own client; the DATEV fees for it are yours, and you keep building in the meantime.

Four things do not move:

• Your record of every contact you wrote. The object cannot be read back, so no layer can reconstruct it. This is the clearest case on this system of a gap that is a fact rather than a feature request.
• The renewal prompt. Re-confirming the connection when its two years are up is your customer’s action, and reminding them is your product’s job.
• The log and the error budget. They run on your infrastructure and are measured against your client, so they cannot be delegated.
• What the advisor changed. Where data travels through Unternehmen Online, the advisor may amend it before it is booked. Reading the ledger back tells you what happened; nothing prevents the amendment.

One note on the objects. Eight combinations here are marked on request, all of them webhooks, and that column is a conversation rather than a switch. Our own data does not support a claim about native events in either direction, so if events on this system matter to your roadmap, the useful move is to ask rather than to infer from the matrix.