Carrying your own reference and routing

Two optional form fields let a CRM / ERP hand over the context Mercura cannot infer from the files:

Field	What it does
`custom_fields`	JSON object of your organisation's custom fields, keyed by the field's label — e.g. `{"Case Number": "00012345"}`. Comes back unchanged as `request_custom_fields` on `GET /tenders/{tender_id}`.
`branch`	Name of the Mercura branch (Niederlassung), matched case-insensitively. Routes the tender to the team that works that branch.

Both are validated against your organisation's configuration before the job is created, so a typo is a plain 400 with no job and no half-created tender: an unknown branch name, an unknown custom-field label, or a duplicate branch name all fail fast. This is deliberately stricter than customer_id — a tender that lands unrouted or without its case number is invisibly wrong, which is worse than a rejected upload.

curl -X POST https://prod-euapi.mercura.ai/api/public/v1/tenders \
  -H "Authorization: Bearer $MERCURA_API_KEY" \
  -H "Idempotency-Key: case-00012345" \
  -F "files=@Leistungsverzeichnis.x83" \
  -F "name=Neubau Bürogebäude Nord" \
  -F "branch=Niederlassung München" \
  -F 'custom_fields={"Case Number": "00012345"}'

Store the request_id that GET /jobs/{job_id} returns: it is the tender_id for every later call, and the stable handle to correlate the tender with the record on your side.

The upload is accepted in seconds; the job stays RUNNING until the processing pipeline finishes (typically a few minutes). Learn the outcome either by polling GET /jobs/{job_id} until status is COMPLETED / FAILED, or by subscribing to the request.processing_completed webhook and correlating by job_id.