5.31 Release Notes - Databricks

Release notes for Harbr version 5.31, covering features and fixes for platforms deployed using an Databricks workspace.

New Features

MCP Discovery Tools

EARLY ACCESS

Harbr's MCP server now exposes a read-only discovery surface, allowing users and AI agents to search, browse, and inspect products and assets on a user's behalf.

• Search products by free text (eg. product name, summary and product description) using prompts related to what you can do with them or by asset type.

• Open a single product to see the caller's own permissions on it, list its assets, and inspect an asset's schema.

• Only what the signed-in user is actually allowed to see is ever returned — the same organisation rules, product visibility, and plan eligibility that apply everywhere else on the platform are respected here too.

Note: this release covers discovery only (search, browse, inspect) — subscribing, exporting, and sharing via MCP are not yet included.

For user guidance, please see https://docs.harbrdata.com/v6/using-the-harbr-mcp . For operator set-up information, see Configuring the MCP Server.

AI Query

Private Preview

AI Query is a new asset type that allows a producer to provision access to an existing Databricks Genie Agent through the Harbr UI. The functionality currently facilitates question-and-answer with text and table responses only.

• Producers register a Genie Agent they've already built by choosing a Databricks connector, selecting from the Genie Agents that connector can reach, and publish.

• Consumers granted the ‘access’ permission on products, are able to ask questions in natural language via a chat modal that opens in place from the product page and My Collection.

• No credentials are shared, no data is moved onto the platform. Rather, every question is proxied through Harbr using the connector's identity, so no user accesses Databricks directly, and each question is recorded for usage and cost visibility.

• Conversations are persisted inside a session. Old sessions are not saved.

For more information on how to create and consume this new asset type, please see https://docs.harbrdata.com/v6/ai-query .

Workload Identity Federation (WIF) for the GCS Connector

Workload Identity Federation (WIF) is now supported for the Google Cloud Storage (GCS) connector. Operators can authenticate to GCS without provisioning or rotating long-lived service-account key files.

• WIF is configured at the connector level during connector creation or update.

• Existing GCS connectors using service-account key authentication continue to function; migration to WIF is optional.

For user guidance and further information, please see Using GCS.

At-Source Cataloguing for Azure Blob Storage

At-source cataloguing for Azure Blob Storage now works even when your storage account sits in a different Microsoft Entra tenant than your Databricks workspace.

Note: This extends the At-Source Tabular Asset Cataloguing for Object Stores feature shipped in 5.30, which noted that cross-tenant configurations were not yet supported for the Azure Blob Storage connector. That restriction has now been lifted.

• Setup is unchanged from your perspective — enter your Azure service principal credentials as normal when creating the connector. Harbr handles the cross-tenant access on its side automatically.

• Existing same-tenant Azure Blob Storage at-source connectors continue to work as before; this adds the cross-tenant case, it does not change the same-tenant path.

For operator detail on how this is provisioned, please see https://ops.harbrdata.com/v5/configuring-the-azure-storage-credential-pool-cas

Delta Table Detection for the Azure Crawler

The Azure connector's crawler now detects Delta tables. The value this enables is that at-source Azure tables in delta format can be added to Delta shares

Note: This extends the At-Source Tabular Asset Cataloguing for Object Stores feature shipped in 5.30. While Delta detection for at-source assets previously worked via the S3 connector, it now also works via Azure Blob Storage.

Improvements

Set-up Improvements for Various Connector Types

Harbr now runs a series of capability checks on a connector before it can be finalized, checking from two vantage points — from Harbr, and from the processing environment — so a non-functional connector is more likely to be identified during setup rather than discovered later.

For user guidance, see the relevant section within https://docs.harbrdata.com/v6/create-a-connector . For operator info, see Configuring Connector Capability Checks.

Desktop Upload Improvements

Desktop asset uploads have been improved for platforms using the upload proxy rather than pre-signed URLs — most relevant where a customer's storage account must remain private.

• Desktop asset uploads no longer time out after four minutes on larger files.

• Fixed a validation bug affecting desktop uploads.

• Finally, binary/desktop-download export zips now build in a way that works on serverless compute.

For operator detail on how to configure this capability, please see Configuring Desktop Upload.

Asset Usage Permission Enforcement

This is the second phase of the asset usage permission changes introduced in 5.30.

The previous release updated what was shown in sharing flows and My Collection, by hiding usage options the platform couldn't fulfil for a given asset or plan. Usage permissions can now only be granted for usage types the asset actually supports — enforcing, at grant time, the same restriction that has been surfaced since 5.30.

Other related improvements to this feature include:

• Asset creators automatically keep full usage permissions, including any new usage types added to the platform later.

• An asset creator's own usage permissions can no longer be changed by other users.

• Usage permissions can no longer be granted on an asset before it has been released.

Note: Each of these three rules is controlled independently via ecosystem metadata and is currently set to off by default. Please see Configuring Usage Permission Enforcement for set-up guidance.

Other Improvements

• New notification type for data-share updates needing attention.

• Connector creation now stops cleanly at the Azure load-balancer limit.

• New connectors list shows full detail parity for every connector type and Connector create/edit now uses a sticky page layout.

• Introduced a scheduled sweep of delta share permissions to ensure accuracy.

• Prevented editing the source database on a connector with 'at source' enabled.

• Databricks connectors now specify their SQL warehouse via a dedicated field, rather than through integration metadata.

• A batch of hardening improvements, including dependency updates and rebuilt images to patch known vulnerabilities, and a fix for an HTML injection issue in the Spaces description field.

• Improvements to platform performance.

‌

Fixes

• Magicbyte validation optimisation for file upload.

• Fixed row-expand control making table rows taller than others.

• Query page can no longer be reached by direct URL when the Query feature is off.

• BigQuery connector now returns a clear message for a malformed service-account key.

• Fixed asset update failures after a prior partial attempt caused by an SFTP folder-creation issue, and intermittent SFTP connection errors during asset read/pull.

• Fixed intermittent timeouts on the Spaces data products endpoint.

• Fixed the Databricks connector silently failing to re-authenticate after its secret was rotated.

‌