Claude Platform OpenTelemetry traces enter public beta

Claude Platform can now send OpenTelemetry traces from the Messages API, Claude Managed Agents, and the Claude SDKs to an observability stack, in public beta. ClaudeDevs said that on October 14, 2026. Managed Agents and the Messages API need a destination in the Claude Console, while Claude SDKs export traces automatically.

Anthropic's article says that if you already send Claude Code or Cowork traces to an observability tool, agents built on the Claude Platform can show up in the same place. Platform docs treat this export as separate from Claude Code, the Agent SDK, and Cowork, which have their own monitoring pages. The article's case for turning traces on is access: review model calls, tool use, and MCP server calls, then widen what an agent can reach as it proves reliable.

A Console destination is the API-tracing path, and it requires an admin. Once that destination is added, the article says, every agent turn, model call, and tool use arrives as a standard OpenTelemetry trace, with latency, tokens, and IDs. SDK tracing is set up in code and needs no extra permission. Docs say API tracing covers Messages API requests, Sessions API requests, token usage, each model run inside a request, Managed Agents turns, and server tools the API runs. SDK tracing covers Messages and Sessions requests, token usage, client tools run by the tool runner, tools in a self-hosted sandbox, and other API requests such as uploading files. Setting up both puts a request's SDK and API spans in one trace. For requests to Claude Platform on AWS, the docs say the API spans are in their own traces instead. API tracing is supported on Claude Platform and Claude Platform on AWS. SDK tracing is not limited to those clouds.

Traces follow the OpenTelemetry GenAI semantic conventions, so Claude data can share tools and dashboards with other providers. The article names Grafana Cloud, Honeycomb, Sentry, and Splunk for sitting agent traces beside logs and metrics, and Arize AI, Braintrust, Datadog, Langfuse, LangSmith, New Relic, and PostHog for turning real traces into evals. Databricks Unity Gateway is the named destination for token usage by team. Any other OpenTelemetry tool that accepts a standard OTLP endpoint works as well. The platform docs list the same partners.

Each destination and each SDK client has a content mode. Metadata only is the default. Spans include identifiers, timing, status, token counts, and text you choose, such as the names and descriptions of agents and tools and the names of MCP servers. Full content adds model prompts and responses. On a Managed Agents session, a full-content tool span also holds the tool arguments and the tool result. If an organization has a HIPAA, FERPA, or K-12 data-handling restriction with Anthropic, destinations export only metadata regardless of content mode. SDK spans are not affected. Full content still works with zero data retention for Messages API requests: the API holds the exported content only in memory until it delivers it. Managed Agents is not eligible for zero data retention, and that retention does not change when OpenTelemetry is on.

With full content, each span holds up to 0.75 MB by default. Over the limit, the SDK or the API shortens the content and sets anthropic.content.truncated to true.

Hyon Lee, CTO at Sterling Labs, said native OpenTelemetry export streams every trace in with no custom instrumentation. Ankush Sharma, head of infrastructure at FloQast, said each model call can sit in the same Grafana Cloud trace as the reconciliation or journal-entry workflow it ran inside. David Traina, senior software engineer at Ramp, said native export from the Claude API can feed an existing improvement loop with no extra steps.

Read the Claude Platform OpenTelemetry docs to choose Console API tracing, SDK tracing, or both, and use Anthropic's product note for the partner guides already published for the tools above.

Sources