Anthropic for Claude API opening OpenTelemetry runtime tracing public beta

AT A GLANCE

Anthropic states that Claude Platform has opened OpenTelemetry runtime tracing public beta, allowing users to view Agent duration, errors, and Token usage.

Article

Anthropic Add Claude API runtime tracing:Agent where it gets stuck and Token where it burns resources can all be investigated

Beating AI news,Anthropic has opened public beta access to the OpenTelemetry tracing feature on Claude Platform. An Agent continuously calls models, searches for information, and executes tools. Developers can now follow execution records to see how long each call took, where errors occurred, and how much Token was consumed. Rather than piecing together the execution process from logs across different services, they can identify the steps that slow tasks down or raise costs.

The most important change this time is native support for API. Claude Code and Cowork previously had similar monitoring capabilities; now Messages API and cloud-hosted Agent can also directly send execution records to platforms such as Grafana, Datadog, and Langfuse. Administrators need only configure the receiving endpoint in Claude Console, without modifying existing API calls one by one. However, tools run by developers themselves still need to supplement tracing records through methods such as SDK.

By default, the system exports only metadata such as duration, status, and Token usage. After enabling Full content, it can also record prompts, model responses, and tool inputs and outputs, facilitating the review of specific errors. However, this content is also sent to external monitoring platforms, so enterprises need to additionally manage access permissions for sensitive data.

Original link https://m.theblockbeats.info/flash/372028

Key points

Anthropic opens the OpenTelemetry tracing feature public beta for Claude Platform.
Execution records can be used to view the duration, errors, and Token usage of each call.
Messages API and cloud-hosted Agent can send execution records to platforms such as Grafana, Datadog, and Langfuse.
Administrators can configure the receiving endpoint in Claude Console without modifying existing API calls one by one.
Tools run by developers themselves still need to supplement tracing records through methods such as SDK.
By default, metadata such as duration, status, and Token usage is exported; after enabling Full content, prompts, model responses, and tool inputs and outputs can also be recorded.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

This feature is like installing queryable records for the Agent execution process. Developers can view the duration, errors, and Token usage of model calls, searches, and tool-execution steps; if Full content is enabled, they can also see more specific input and output content.

Why it matters to readers

It may help developers identify steps that slow tasks down or raise costs, and connect execution records to monitoring platforms such as Grafana, Datadog, and Langfuse.

If you use Claude API to build Agent, this can help you understand which step a task has reached, where errors occurred, and Token consumption.

Risks and unknowns

Evidence not provided Anthropic No official announcement was provided; the scope of the public beta and formal release date have not been officially confirmed.
After enabling Full content, prompts, model responses, and tool inputs and outputs will be sent to external monitoring platforms, and access permissions for sensitive data require additional management.
The evidence does not specify the actual tracing fields, fees, data retention period, data-masking method, or access-control details.

Anthropic for Claude API opening OpenTelemetry runtime tracing public beta

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center