The Pulse
Anthropic Adds OpenTelemetry Tracing to Claude Agents
Anthropic announced OpenTelemetry tracing for Claude Managed Agents, the Messages API and Claude SDKs on October 14, 2026. The public beta exports agent turns, model calls, tool use, timing and token data to observability platforms, with pr
AI.info Team · Written with AI ·
Anthropic announced on October 14 that developers can trace Claude Managed Agents, Messages API requests and Claude SDK activity through OpenTelemetry, the standard used to send telemetry data to observability systems. The public beta records agent turns, model calls and tool use, letting teams inspect timing, token usage and request identifiers alongside other application data.
API tracing can start in the Console
For the Messages API and Managed Agents, an organization administrator adds an observability destination in the Claude Console. Anthropic says the API can then emit spans for each model request and each tool use without developers changing their application code. A destination specifies the provider’s OTLP endpoint and authentication details; the API sends traces for the workspaces assigned to it.
SDK tracing works differently: developers register a tracer provider in their application, and the SDK sends spans through the application’s OpenTelemetry exporter. Teams can enable API tracing, SDK tracing or both. Combining them can place SDK and API spans in one trace, though requests made through Claude Platform on AWS keep their API spans in separate traces.
Traces can include prompts—or leave them out
Metadata-only mode is the default. It records identifiers, timing, status and token counts, along with selected text such as agent and tool names. A separate full-content mode adds prompts and model responses, which can help teams investigate slow or failed runs and build evaluations from real traffic.
That extra visibility also sends conversation content to the chosen observability provider. Anthropic’s documentation says the provider holds exported data under its own access and retention settings, so organizations need to review those controls. For organizations with HIPAA, FERPA or K-12 data-handling restrictions, API destinations export metadata only regardless of the selected content mode; SDK spans are not affected by that restriction.
One format, several observability platforms
Anthropic says the traces follow OpenTelemetry’s GenAI semantic conventions, allowing compatible tools to recognize shared attributes. The company names Grafana Cloud, Honeycomb, Sentry and Splunk for viewing agent traces alongside logs and metrics, and Arize AI, Braintrust, Datadog, Langfuse, LangSmith, New Relic and PostHog for working with production traces and evaluations. Other platforms can receive the data if they accept an OTLP endpoint.
Anthropic’s announcement frames tracing as a way to inspect agents before granting them broader access. Foxglove, a company building agents for robotics teams, says the export lets it analyze agent interactions in PostHog alongside product usage data. “That helps us evaluate how our agents perform in practice, identify where they fall short, and prioritize improvements for robotics teams,” said Roman Shtylman, Foxglove’s co-founder and CTO.
Beta status puts data handling in view
The feature remains in public beta, and Anthropic warns that beta terms apply. The distinction between API-side and SDK-side tracing also matters for deployment: API tracing covers any client using the platform, while SDK tracing covers clients using Anthropic’s SDK. API destinations require administrator permissions, even as developers can view configured destinations.
For teams enabling full-content traces, the destination becomes part of the data-handling boundary: prompts and responses leave Anthropic’s platform for the provider they select. Anthropic’s documentation makes that choice explicit, with metadata-only export as the default and full conversation capture available as an opt-in setting.