DATEV and HubSpot integration

What moves between them

The flow runs one direction, from HubSpot into DATEV. ml-connector reads invoices and their line items from HubSpot, along with the associated company and contact, and submits them into DATEV as booking jobs: an EXTF CSV booking batch for the finalized entry, or a DXSO XML job to DATEV Unternehmen Online for a booking suggestion. The invoice PDF is uploaded to DATEV Unternehmen Online through the documents API against the right client-specific document type. Companies and contacts are written into the EXTF master data file as creditor and debtor records. Nothing financial flows back from DATEV, because EXTF bookings are write-only and DATEV exposes no posted journals or GL accounts to read.

How ml-connector handles it

ml-connector stores both credential sets encrypted. On the HubSpot side it sends the Private App token as a Bearer header and pages invoices and line items with limit and after. On the DATEV side it runs the OAuth 2.0 PKCE login once with the tax advisor, then refreshes the 15-minute access token automatically, sending only the client id on refresh as DATEV requires. It first calls accounting:clients to resolve the DATEV client id used as a path parameter on every other call, and fetches the client-specific document types before any PDF upload rather than hardcoding them. HubSpot GL accounts and tax codes do not exist, so the DATEV account and Steuerschluessel for each booking line come from a mapping you configure once, since DATEV will not return its chart of accounts. EXTF CSV is written as NFC-normalized UTF-8 with a whitelisted filename, because non-precomposed characters are silently rejected. DATEV processing is asynchronous and gives no webhook, so ml-connector polls each job id with exponential backoff and jitter until it reports complete or failed. Duplicate submissions are avoided by generating stable, deterministic EXTF filenames, which DATEV treats as the dedup key, alongside a BullMQ job id, so a re-read HubSpot invoice is never booked twice. Document uploads use the idempotent PUT with a supplied GUID over POST. Every record carries a full audit trail and can be replayed if a job fails.

A real-world example

A German digital agency with around 60 staff runs HubSpot as its CRM and raises client invoices there, while its external tax advisor keeps the books in DATEV. Before the integration, an office manager exported each HubSpot invoice and emailed the PDFs to the advisor, who re-keyed the bookings into DATEV and chased mismatched customer names and missing receipts during every monthly close. With DATEV and HubSpot connected, each HubSpot invoice is submitted as a DATEV booking job and its PDF uploaded to DATEV Unternehmen Online within the sync window, with the customer written as a debtor record and the line mapped to the agreed GL account. The advisor opens a clean booking batch instead of a folder of emails, and the manual re-keying step is gone.

What you can do

• •Submit HubSpot invoices and line items into DATEV as EXTF CSV booking batches or DXSO XML jobs.
• •Upload HubSpot invoice PDFs to DATEV Unternehmen Online against the correct client-specific document type.
• •Write HubSpot companies and contacts into DATEV as creditor and debtor master data.
• •Bridge the HubSpot Private App Bearer token and DATEV's OAuth 2.0 PKCE login with automatic token refresh.
• •Poll each DATEV job to confirmation with stable-filename dedup, retries, and a full audit trail on every record.

Questions

Which direction does data move between DATEV and HubSpot?

The flow is one direction, from HubSpot into DATEV. Invoices, line items, and the companies and contacts behind them move from HubSpot into DATEV as booking jobs and uploaded documents. DATEV bookings submitted as EXTF files are write-only and DATEV exposes no readable chart of accounts, so ml-connector never writes financial data back into HubSpot.

How does ml-connector handle DATEV having no webhooks?

DATEV does not push any events, and all booking and document imports are asynchronous. After ml-connector submits an EXTF or DXSO job it receives a job id and polls the DATEV status endpoint with exponential backoff and jitter, from about 5 seconds up to a 60-second cap, until the job reports complete or failed. The HubSpot side is read on a schedule the same way, since a Private App token cannot receive HubSpot webhooks.

What maps to what, given DATEV does not expose GL accounts?

HubSpot invoices and line items map to DATEV booking rows, and HubSpot companies and contacts map to DATEV creditor and debtor master data in the EXTF file. Because DATEV will not return its chart of accounts through the API and HubSpot has no GL accounts at all, the DATEV account number and tax code for each line come from a mapping you configure once. ml-connector validates that mapping before submitting a job so postings land on valid accounts.