Fabric's Salesforce connector will not point at a host you run
note
We were planning a Salesforce load into Fabric and assumed the Salesforce objects connector could be pointed at a host we control, either a test double or an adapter in front of the real org. It cannot, and the reason is stated on the page if you read it as a rule rather than as a note.
What the pages say
Three Learn pages cover the connector: the connection setup page, the overview, and the copy activity page. Between them:
- There is exactly one authentication kind, "Organizational account", for Dataflow Gen2, the pipeline copy activity and the copy job alike. It is a browser sign-in against Salesforce. No client credentials, no service principal, nothing a scheduled job can hold.
- The custom URL field is a host allowlist, not a free-form address: "Custom URLs only support the salesforce.com and cloudforce.com domains. The use of other domains results in an 'Invalid Arguments' error." The same page permits a CNAME that redirects to Salesforce, which is Salesforce's own custom-domain feature, where Salesforce still terminates the name.
Where the confusion comes from
The shape people describe when they say the connector takes a URL, a token endpoint and a query, is Azure Data Factory's Salesforce V2 linked service. That one does take a free-form environment URL and OAuth client credentials. It is a different product with a similar name, and searching for one turns up the other.
What we did instead
We wrote the extractor. Salesforce's Bulk API 2.0 query surface is four calls, and owning it means the double we already had becomes the extractor's fixture rather than something Fabric has to be talked into accepting. Fabric's generic REST and HTTP connectors would also have worked and take any host.
Read the pages before relying on this: connector capabilities move, and this was true on 2026-09-05.