Databricks SDK releases added workspace services for AI functions, sandboxes, and MCP credentials.

September brought major service additions across both the Python and Go SDKs. Databricks introduced dedicated workspace-level client interfaces for AI Functions, Domains, and Sandbox environments [13], [15], [16], [17], followed later in the month by the Mason service client [6]. AI Gateway tooling expanded considerably; releases added support for Model Context Protocol (MCP) service credentials [7], [8] and introduced programmatic skill management methods to create, list, update, and finalize skills [3]. The month closed with the addition of the ai_decide method for AI functions [1].

SDK coverage for data pipelines and feature engineering also broadened. Pipelines gained optional source tables and schemas to better accommodate streaming connectors [17], along with ingestion support for TikTok Ads, Smartsheet, Avro, and Protobuf formats [7], [8]. Feature engineering interfaces received backfill management operations and a purge method for feature entities [13], [14]. Practitioners had to adjust client code to handle breaking changes that stripped legacy routing and rate-limiting configuration fields from model serving APIs [13], [16].

Client configuration and platform integration saw notable operational additions. The Go SDK introduced custom HTTP header configuration via Config.Headers and the StaticHeaders helper [2]. Job configurations across the board gained environment variable support for jobs, runs, and tasks [1]. In external tooling, MLflow 2.11.5 introduced an option to route Unity Catalog model registry artifact transfers directly through the Databricks SDK Files API [5].