Claude Platform can now send native OpenTelemetry traces to Databricks MLflow

Teams that build agents on Claude can now send execution traces straight into Databricks MLflow without writing adapter code. Databricks announced the integration in a blog post on November 3, and DeveloperTech News reported on November 4 that Anthropic had added Databricks MLflow as a supported destination for the Claude Platform's native OpenTelemetry export.

What shipped

Anthropic introduced native OpenTelemetry (OTel) export on October 14 as a public beta. The Claude API emits a span for each model request and tool call, and an organisation admin adds a "destination" in the Claude Console pointing at any observability backend that accepts OTLP over HTTPS. At launch, Anthropic listed partners such as Datadog, Grafana Cloud, Honeycomb and Langfuse; the only Databricks product it named was Unity Gateway, for usage tracking.

The new integration covers the three surfaces that use this export: the Claude SDKs, the Messages API and Claude Managed Agents. Claude Code and the Claude Agent SDK can send traces to the same place with extra environment variables. On the Databricks side, spans are received by Zerobus Ingest and written to Unity Catalog tables tied to an MLflow experiment, so access is governed by Unity Catalog and the data can be queried in SQL.

Databricks describes a three-step setup: create an MLflow experiment that stores traces in Unity Catalog, configure the export with a workspace ID, region, spans table and a table-scoped OAuth token, then run an agent and inspect the result. Prompts and outputs are only captured if content mode is set to full; otherwise only metadata is recorded.

What it is for

Once traces are in MLflow, Databricks points to several tools: an "Analyze trace" action that uses its Genie Code assistant to explain where a run failed, MLflow Issues (in beta) to cluster recurring problems across traces, review queues for domain experts, and Agent Evaluation, which applies LLM judges and compares baseline and revised agents side by side.

Limits

The Claude export itself is still a public beta, and both companies describe capabilities rather than published performance figures. The Databricks post does not specify supported clouds or regions, and says setup requires certain permissions and a SQL warehouse without detailing them in the post itself. Anthropic's documentation warns that once spans are delivered, they are held by the chosen service under its own retention and access rules.