# Claude Platform traces

## Start tracing in 5 minutes

Send OpenTelemetry traces of your Claude API requests and Managed Agents sessions to Better Stack. Each model request, tool use, and agent turn arrives as a span, with token usage, latency, and errors ready for dashboards.

### 1. Create a Claude Platform source

Create a [new Claude Platform source in Better Stack](https://telemetry.betterstack.com/team/0/sources/new?platform=claude_platform&name=Claude%20Platform ";_blank")

### 2. Add a destination in the Claude Console

As an admin of your Claude Console organization:

1. Open the [OpenTelemetry settings page](https://platform.claude.com/settings/telemetry-destinations ";_blank") and click **Add destination**.
2. Set **Provider** to **Custom**.
3. Set **OTLP endpoint URL** to the URL below. Don't add `/v1/traces`. Claude adds it.
4. Under **Request headers**, add the header `Authorization` with the value below.
5. Keep **Content mode** at **Metadata only**. Choose **Full content** to also send prompts and responses.
6. Set **Workspaces** to **All**.
7. Click **Add destination**.

```
[label OTLP endpoint URL]
https://$INGESTING_HOST
```

```
[label Authorization header value]
Bearer $SOURCE_TOKEN
```

### 3. Send a test span

1. Click **Test** in the destination's **More actions** menu.
2. Open [Live tail](https://telemetry.betterstack.com/team/0/tail ";_blank") and select your Claude Platform source marked as `spans`.
3. Check that the `anthropic.test` span arrived.

[info]
#### Test failed?
The Console's message names the error. HTTP 401 or 403 means the `Authorization` header value doesn't match your source token.
[/info]

### 4. Start tracing

Messages API requests and Managed Agents sessions in your workspaces now send their spans to Better Stack, under the service `claude-platform`. See [Tracing](https://betterstack.com/docs/logs/tracing/) for how to explore them.

## Chart tokens and errors on dashboards

The source extracts these metrics from the spans, so you can chart them in [dashboards](https://betterstack.com/docs/logs/dashboards/):

- **Token usage**: `gen_ai_input_tokens`, `gen_ai_output_tokens`, `gen_ai_cache_read_input_tokens`, and `gen_ai_cache_write_input_tokens`. Each token is counted once, following [Anthropic's counting rule](https://platform.claude.com/docs/en/manage-claude/opentelemetry#count-tokens). Input tokens already include cache reads and writes.
- **Labels**: `gen_ai_model`, `gen_ai_operation`, `gen_ai_tool_name`, `gen_ai_agent_name`, `error_type`, `stop_reason`, `workspace_id`, and `outcome_evaluation_result`.
- **Span count and latency**: `duration`, as on any traces source.

For every span and attribute Claude sends, see [Anthropic's span reference](https://platform.claude.com/docs/en/api/opentelemetry/spans).

## Need help?

Please let us know at hello@betterstack.com.
We're happy to help! 🙏