Databricks MLflow has become a primary destination for native OpenTelemetry trace exports from the Claude Platform. This integration allows developers to leverage unified tracing for agent executions, combining detailed step-level telemetry with enterprise-grade data governance and scalable ingestion through Unity Catalog and Zerobus.
- Native OpenTelemetry tracing for Claude agents integrated with Databricks MLflow
- Unity Catalog ensures secure, governed access to sensitive trace data
- MLflow’s tools enable detailed debugging and recurring issue detection
Infrastructure signal
The new native export capability from Claude Platform to Databricks MLflow leverages OpenTelemetry, an open standard for telemetry data, to record and send detailed execution traces. These traces consist of spans representing individual steps such as model invocations and tool interactions. The spans are ingested through Zerobus, which supports high query-per-second (QPS) rates and low latency, ensuring scalability for enterprise workloads.
Trace data is persisted in Unity Catalog tables, providing centralized governance and fine-grained access control for sensitive information contained in agent traces. This architecture supports SQL querying of trace data, enabling data teams to integrate observability signals with existing analytics workflows and compliance requirements.
Developer impact
Developers using Anthropic SDKs, the Messages API, or Managed Agents can now instrument their agents with native OpenTelemetry tracing and export directly to MLflow, simplifying observability setup. This streamlined tracing enables in-depth inspection of agent behavior including model calls, timing, tool usage, and optionally captured inputs and outputs for debugging and performance tuning.
MLflow enhances debugging with Genie Code-powered analysis, automatically surfacing root causes for unexpected agent outputs, like missing context or failed tool invocations. The platform also supports scheduled detection of recurring operational or quality issues, allowing teams to monitor and track fixes over time, which improves reliability and iterative agent development.
What teams should watch
Teams focused on agent observability and reliability should adopt the native OpenTelemetry integration to unify tracing across Claude agents and other instrumented services. Monitoring trace ingestion rates, query performance in Unity Catalog, and the accuracy of issue detection will be critical to optimizing cloud costs and observability value.
Security and compliance teams must leverage Unity Catalog’s governance features to control access to sensitive trace data appropriately. Additionally, Development teams should evaluate how enabling full trace capture of inputs and outputs affects privacy policies and storage requirements. Close coordination between cloud infrastructure, developer productivity, and data governance groups will maximize benefit from this new integrated platform capability.