Claude Platform adds OpenTelemetry trace export for agents, API and SDKs

Anthropic put OpenTelemetry trace export for Managed Agents, the Messages API and Claude SDKs into public beta, sending spans to any OTLP backend.

Anthropic has opened a public beta that sends OpenTelemetry traces from Claude Managed Agents, the Messages API and the Claude SDKs to any tool that accepts a standard OTLP endpoint, according to its October 14 announcement. Each agent turn, model call and tool use arrives as a span with latency, tokens and IDs. Anthropic says the spans follow the OpenTelemetry GenAI semantic conventions, so Claude data can sit in the same dashboards as other providers'. For anyone running agents, it means you can see which step failed or slowed down without writing your own instrumentation.

There are two setup paths, per Anthropic's docs. SDK tracing is configured in code by registering a tracer provider and needs no special permission. API tracing needs an Admin to add a destination in the Claude Console. It also captures things the SDK can't see, such as each model run within a request, Managed Agents turns and server tools. Set up both and a request's SDK and API spans land in one trace.

The docs say content mode defaults to Metadata only: identifiers, timing, status and token counts. Full content adds prompts and responses, capped at 0.75 MB per span by default. Organizations with HIPAA, FERPA or K-12 restrictions get metadata only from Console destinations.

Vendors are already building on it. New Relic says Bedrock and Vertex AI support is coming soon. MLflow published a walkthrough using its open-source server as the backend. One catch from that guide: Console export can't reach a localhost server, so you need a reachable HTTPS endpoint. TokenPost reports that Anthropic hasn't given pricing or a GA timetable.

Sources