Claude Platform Adds Native OpenTelemetry Export for Agent Observability

Anthropic has enabled native OpenTelemetry (OTel) export on the Claude Platform, allowing developers to send traces from Claude Managed Agents, the Messages API, and Claude SDKs to any OTel-compatible tool. This integration lets teams follow every step an agent takes, including model calls, tool use, and MCP server interactions, directly within their existing observability stacks.

Integration and Standards

The traces adhere to the OpenTelemetry GenAI semantic conventions, ensuring that data from Claude interoperates seamlessly with data from other providers in the same dashboards. By adding a destination in the Claude Console, every agent turn arrives as a standard OTel trace containing latency, token counts, and IDs. This approach allows developers to debug agents alongside the rest of their infrastructure, identifying whether a slow or failed step originated from the model, a tool, or an external service.

Ecosystem Support

The feature supports a wide range of observability and AI engineering platforms:

• Debugging and Monitoring: Integrations are available for Grafana Cloud, Honeycomb, Sentry, and Splunk, placing agent traces next to logs and metrics.
• Evaluation and Improvement: Tools like Arize AI, Braintrust, Datadog, Langfuse, LangSmith by LangChain, New Relic, and PostHog can use these traces to create evals, catch quality issues, and measure the impact of changes.
• Organizational Tracking: Databricks Unity Gateway can ingest traces to help platform and security teams monitor token usage by team and flag unusual activity.

Availability and Early Results

The feature is currently in public beta. While Claude Managed Agents and the Messages API require configuring a destination in the Claude Console, Claude SDKs export traces automatically. Early adopters report significant benefits; for instance, Ramp noted that using traces to feed their internal improvement system helped cut token usage by 10–20% and make sessions 30% faster on some agents. FloQast highlighted the ability to answer audit questions by viewing prompt versions, tool calls, and latency in a single timeline alongside their reconciliation workflows.