Databricks MLflow is now a supported destination for native OpenTelemetry export from the Claude Platform, allowing developers using the Anthropic SDKs, the Messages API and Managed Agents to send native traces to Databricks and use MLflow to debug their agents and evaluate changes. MLflow Tracing is built on OpenTelemetry, and Claude’s native instrumentation generates spans such as model requests and tool calls and exports them using the OpenTelemetry Protocol. Databricks receives the spans through Zerobus Ingest and stores them in Unity Catalog tables linked to an MLflow experiment, so Claude traces can be inspected alongside traces from other instrumented services, with Unity Catalog controlling access to the data, which can also be queried with SQL. The combination of Zerobus and Unity Catalog provides low-latency trace ingestion at high QPS plus the ability to govern access to sensitive data stored in agentic traces. Setup involves creating an experiment in Databricks with Unity Catalog selected as the trace storage location, configuring native export using workspace ID, region, and spans table, and running a task and inspecting the trace; inputs and outputs are captured only if content mode is set to “full.”