Agent frameworks
Claude Agent SDK tracing for Python
Out-of-process agent runs are still traced: the wrapper reads tokens, cost and latency from the ResultMessage.
Set up Claude Agent SDK
The snippet is copied from the SDK README, not paraphrased. Keys shown as tk_live_... are the environment key from Settings › Environments.
import evalkit
evalkit.init(
subscription_key="tk_live_...", # Dashboard → Settings → Tracing
service_name="my-service",
)Copied from sdk-py/README.md § Quick start
Install → first trace
- 01
Install the SDK
pip install syntropylabs-evalkit or npm install syntropylabs-evalkit.
- 02
Call init() once
Pass your environment key and a service name, as early as possible in the process, before other modules run requests.
- 03
Run one query() with the Claude Agent SDK
No wrapper and no decorator: the call is traced automatically with model, latency and tokens.
- 04
Open the first trace
Traces shows the call as a span; the environment’s quickstart page ticks when it arrives.
What a turn looks like
Sample data in the product’s own table. Hover, focus or tap a claim to see the columns it points at.
Sample run
| Operation | Service | Status | Model | Latency | Tokens | ≈Cost | Score |
|---|---|---|---|---|---|---|---|
| support_agent.turn8f3a1c0d94e2session sess_4b1e | support-api | ok | gpt-4o-mini | 1.84 s | 3,412 | $0.0006 | 92%auto |
| refund_agent.turnc21d7e5a30b8session sess_9a02tool loop | refund-worker | ok | claude-sonnet-4 | 4.31 s | 7,905 | $0.0389 | 68%auto |
| claude-code.turn5be04f7d1a96session sess_f77c | Claude Code | unset | claude-sonnet-4 | 48.20 s | 61,208 | $0.19 | — |
| rag.answere9a2b6c4d015error | docs-bot | error | gpt-4o | 6.02 s | 9,880 | $0.0312 | 41%auto |
| voice.turn17c8d3f2a4e0session sess_20d1 | ivr-agent | ok | gpt-4o-realtime | 0.92 s | 1,104 | $0.0071 | — |
fig. 1 · the traces table this integration fills · sample data
What you see
Taken from the SDK READMEs’ coverage tables. with flag means the signal leaves your machine only when you opt in.
| Signal | Captured | Detail |
|---|---|---|
| query() and ClaudeSDKClient.receive_response() as spans | yes | automatic via init() when claude_agent_sdk is installed |
| Tokens, cost and latency | yes | read from the ResultMessage |
| Time to first token on streams | no | the model call runs in a subprocess the in-process patch cannot see |
| Prompts and completions | yes | on by default; capture_content=False / captureContent: false keeps every metric and drops every payload |
| Your own functions | yes | every function in your app’s source tree, on by default |
Limitations
- Python only.
- If claude_agent_sdk is installed after init(), call evalkit.patch_claude_agent_sdk() explicitly.
- Individual model calls inside the subprocess are not visible; the span covers the query and what the ResultMessage reports.
Next steps
Trace Claude Agent SDK today
Tracing is free for one project on every plan, coding-agent traces included. Evaluations, datasets and simulation are Pro and up.