Skip to content

v1.3.1 — Mason attribution in Databricks audit logs

Choose a tag to compare

@DoyleDev DoyleDev released this 06 May 04:28
· 113 commits to main since this release

What's new

Patch release — internal attribution for the ai-dev-kit MCP integration.

Changes

  • The ai-dev-kit MCP server now spawns with DATABRICKS_SDK_UPSTREAM=mason and DATABRICKS_SDK_UPSTREAM_VERSION=1.3.1 set on its env. The Databricks Python SDK appends those to its User-Agent header (... upstream/mason upstream-version/1.3.1), so warehouse audit logs (system.access.audit.request_params.user_agent) can attribute MCP-driven SQL back to Mason.
  • Dropped DATABRICKS_CONFIG_PROFILE from the MCP entry's env. Profile changes in Mason now propagate to the SDK without requiring an MCP restart — the SDK falls through to its default profile resolution chain (~/.databrickscfg, DATABRICKS_HOST/DATABRICKS_TOKEN, etc.).
  • One-time migration on app start patches any pre-existing ai-dev-kit MCP entry whose env still includes the old DATABRICKS_CONFIG_PROFILE or lacks the upstream vars.

The "Source" column in query history will continue to show Databricks-ai-dev-kit (that's ai-dev-kit's product token, set in their Python code and not overridable from outside). The Mason attribution is in the full User-Agent string and visible via:

SELECT request_params.user_agent
FROM system.access.audit
WHERE request_params.user_agent LIKE '%upstream/mason%';

Install (macOS, Apple Silicon)

curl -fsSL https://raw.githubusercontent.com/databricks-solutions/mason/main/scripts/install.sh | bash

Or download Mason-1.3.1-arm64.dmg below. Build is signed with a Developer ID and notarized.