Okta + Databricks integration
Set up Okta as the identity provider for Databricks via SAML + SCIM — across both workspace-level + account-level Databricks identity.
- Okta admin
- Databricks account admin
- Databricks Unity Catalog enabled (recommended)
1. Create a new SAML 2.0 application in Okta
In the Okta admin console, create a new SAML 2.0 application. Choose "Web Application" type. Note the placeholders for ACS URL + Entity ID — you'll get these from Databricks in step 3.
2. Get the SAML metadata URL from Okta
Okta exposes the IdP metadata at a stable URL. Copy this URL — you'll paste it into Databricks's SSO configuration. Alternatively, download the metadata XML if Databricks doesn't support URL-based metadata.
3. Configure SSO in Databricks
In Databricks's admin → security → SSO settings, paste the Okta metadata URL (or upload the XML). Databricks will display the ACS URL + Entity ID it expects — copy these.
4. Return to Okta + complete the SAML app config
Paste Databricks's ACS URL into the Okta app's Single Sign-On URL field. Paste the Entity ID into the Audience URI field. Set the NameID format to EmailAddress (or persistent if Databricks expects that).
5. Configure attribute mapping
Map the attributes the SP expects (see the Attribute Mapping section below). At minimum, email is required. Most apps also expect firstName + lastName.
6. Assign users + groups
In Okta, assign the SAML app to users or groups that should have access. Test with a pilot group before broad rollout.
7. Test end-to-end
Sign in to Databricks via the IdP-initiated link (from Okta dashboard) AND via SP-initiated (direct Databricks login URL). Both should work. Check the SAML Tracer browser extension or SAML decoder to inspect the assertion if anything fails.
What flows from where.
| Source (Okta) | Target (Databricks) | Note |
|---|---|---|
| user.email | NameID | — |
| user.email | userName | — |
| user.firstName | givenName | — |
| user.lastName | familyName | — |
- Clock skew: Okta and Databricks clocks must be within ~5 minutes. NTP-sync both. SAML's NotBefore + NotOnOrAfter are strict.
- NameID format mismatches are the most common failure. Databricks typically wants EmailAddress; Okta defaults vary. Mismatch → cryptic "invalid assertion" errors.
- Just-in-time (JIT) provisioning vs SCIM: many apps support both. SAML JIT creates the user on first SSO; SCIM creates them ahead of time. Pick one — both can cause attribute drift.
- Audience restriction: Databricks's expected Audience URI must match exactly what the IdP sends. Trailing slashes + protocol (http vs https) matter.
- Signed Response vs signed Assertion: many SPs require the Assertion to be signed (not just the Response envelope). Check the SP's docs.
- Databricks has account-level + workspace-level identity. Modern setups use account-level SCIM with workspace assignment.
- Service principals are managed separately from human users. Don't SCIM-provision SPs.
- Unity Catalog identity flows differently — make sure SCIM provisioning targets the account console, not individual workspaces.
- IdP-initiated SSO works (sign in from the IdP dashboard)
- SP-initiated SSO works (visit Databricks directly + get redirected to IdP)
- User attributes flow through correctly (email, name, groups)
- Logout (single logout if supported) works as expected
- Step-up MFA fires when policy requires it
- Unauthorized users (not assigned to the app) get a clean denied message
- Capture a successful SAML response and inspect it (use the SAML decoder tool)
For the latest vendor-side configuration changes, refer to:
Databricks SAML + SCIM →