Anthropic launches OpenTelemetry export in beta for Claude Platform: Messages API and Managed Agents traces go to your own observability stack

Why it matters

【Key points】

On Oct 14 Anthropic’s Claude Platform release notes launched OpenTelemetry export in beta. Messages API requests and Claude Managed Agents sessions can be traced into your own observability backend. The Claude SDK emits spans in Python, TypeScript, C#, Go, Java, PHP and Ruby; API-side spans need an OpenTelemetry destination in the Claude Console.

You can enable SDK tracing, API tracing, or both so client and server spans land in one trace. Coverage includes Messages/Sessions requests, token usage, model runs inside a request, Managed Agents turns, and server or client tools depending on which side you enable. Content mode defaults to metadata_only; Full content can put prompts and responses in spans (destinations stay metadata-only under HIPAA-style restrictions). Partner OTLP targets listed include Arize, Braintrust, Datadog, Grafana, Langfuse, LangSmith, Sentry and Splunk.

@ClaudeDevs posted the same day (~98k views, 1.7k+ likes).

【Verifiable】

Docs: https://platform.claude.com/docs/en/manage-claude/opentelemetry

Release notes (Oct 14, 2026): https://platform.claude.com/docs/en/release-notes/api

X: https://x.com/ClaudeDevs/status/2110421732029063352

【Our read】

1. This moves Claude from opaque billing into stacks teams already run. Multi-agent tool chains are hard to debug from logs alone; OTel plus GenAI semantic conventions cuts the cost of wiring Datadog or Grafana.

2. Beta schema can still change—pin the SDK. Full content ships prompts off-box to the observability vendor’s retention policy; metadata_only is the safer default.

3. Export is best-effort, not a complete usage or leak detector; the docs still point to the Usage API for full counts.