Claude Agent SDK

All Claude Agent SDK integration capabilities

About

Ship telemetry from agents you build on the Claude Agent SDK to Sazabi. The SDK runs Claude Code under the hood, so every agent run emits the same structured events as a Claude Code session — user prompts, tool calls, tool results, and model API requests with token counts and cost — with no extra instrumentation library. Events carry app.entrypoint set to sdk-py or sdk-ts, so you can tell production agents apart from interactive Claude Code sessions. claude_code.api_request events carry model, input_tokens, output_tokens, cache_read_tokens, cost_usd, and duration_ms; every event carries session.id.

Setting OTEL_LOG_USER_PROMPTS=1 exports the raw prompt text your agent sends, which in production often includes your end users’ input. The snippets below keep it off; token counts, cost, and tool calls are exported either way.

Prerequisites

An agent built on the Claude Agent SDK for Python or TypeScript
Your Sazabi intake URL, shown above — open the log source and copy the Intake URL from its setup screen (the public key is embedded in the URL)

Set up in the dashboard

Store the intake URL with your agent’s configuration

Make the intake URL available to your agent as an environment variable named SAZABI_INTAKE_URL — in your deployment’s secret store, or your shell for a local run. Keep it out of committed source.

Pass the telemetry variables in your agent options

Add the following variables to the env option of every query() call or ClaudeSDKClient your agent creates.

Python
TypeScript

agent.py

import os
from claude_agent_sdk import ClaudeAgentOptions, query

options = ClaudeAgentOptions(
    env={
        "CLAUDE_CODE_ENABLE_TELEMETRY": "1",
        "OTEL_LOGS_EXPORTER": "otlp",
        "OTEL_EXPORTER_OTLP_PROTOCOL": "http/protobuf",
        "OTEL_EXPORTER_OTLP_ENDPOINT": os.environ["SAZABI_INTAKE_URL"],
        "OTEL_METRICS_INCLUDE_ENTRYPOINT": "true",
        "OTEL_LOG_USER_PROMPTS": "0",
        "OTEL_LOG_ASSISTANT_RESPONSES": "0",
    },
)

async for message in query(prompt=prompt, options=options):
    ...

In Python, env is merged over your process environment.

agent.ts

import { query } from "@anthropic-ai/claude-agent-sdk";

const telemetryEnv = {
  ...process.env,
  CLAUDE_CODE_ENABLE_TELEMETRY: "1",
  OTEL_LOGS_EXPORTER: "otlp",
  OTEL_EXPORTER_OTLP_PROTOCOL: "http/protobuf",
  OTEL_EXPORTER_OTLP_ENDPOINT: process.env.SAZABI_INTAKE_URL,
  OTEL_METRICS_INCLUDE_ENTRYPOINT: "true",
  OTEL_LOG_USER_PROMPTS: "0",
  OTEL_LOG_ASSISTANT_RESPONSES: "0",
};

for await (const message of query({ prompt, options: { env: telemetryEnv } })) {
  // handle messages
}

In TypeScript, env replaces the environment the SDK passes to Claude Code. Keep the ...process.env spread so your agent still sees ANTHROPIC_API_KEY and PATH.

Keep OTEL_METRICS_INCLUDE_ENTRYPOINT at true — without it, events carry no app.entrypoint. Leave the metrics exporter off to keep volume down; if you enable it, Sazabi stores each metric data point as a log record.

Redeploy your agent

The SDK reads the variables each time it starts Claude Code. Deploy the change (or restart your agent locally), then run one query.

Set up with the CLI

You can also register the Claude Agent SDK log source with the Sazabi CLI (installed and authenticated — see CLI reference). Registering the source mints the same intake URL the dashboard shows above (the public key is embedded in its hostname):

sazabi log-sources create claude_agent_sdk --mode manual

The command prints the intake URL for the new source — copy the whole URL and point your agent at it using the configuration shown above. Run sazabi log-sources get <log-source-id> at any time to reprint the endpoint, or sazabi log-sources list to see every log source in the project.

Verify

Run one query through your agent. Open the Intake page in the Sazabi dashboard and confirm events from the Claude Agent SDK source appear within a minute or two. From the CLI,

sazabi logs ask "How many claude-code log lines with app.entrypoint sdk-py or sdk-ts arrived in the last 15 minutes, by event name?"

confirms events are arriving. Start a thread and search for claude_code.api_request to see each model request with its token counts and cost, or for claude_code.api_error to find failed requests.

Troubleshooting

No events after a run — Confirm SAZABI_INTAKE_URL is set in the process that runs your agent and that the env option reaches the query() call or client that ran. Claude Code sends its batch when the run ends; a process killed mid-run may not send it. TypeScript agent can no longer reach Anthropic — The env object replaced the whole environment. Restore the ...process.env spread. Events appear without app.entrypoint — OTEL_METRICS_INCLUDE_ENTRYPOINT is missing or not true. Set it and redeploy. Events appear under service claude-code — That is expected: the SDK runs Claude Code, which names its own service. Filter on app.entrypoint to separate agents from interactive sessions, or add OTEL_RESOURCE_ATTRIBUTES with a deployment.environment= entry to tell environments apart. Metric data points appear as log records — Expected when OTEL_METRICS_EXPORTER is set: Sazabi stores each metric data point as a log record, not in a metrics store. Unset it to keep volume down; api_request events already carry token counts and cost.

Claude Code

Ship interactive Claude Code session telemetry.

Intake

See per-source ingestion volume and health.

Log sources

Browse the full log source catalog.

Overview

Get started

Guides

Catalogs

Reference

Security & trust

cve.now

Claude Agent SDK

About

Prerequisites

Set up in the dashboard

Store the intake URL with your agent’s configuration

Pass the telemetry variables in your agent options

Redeploy your agent

Set up with the CLI

Verify

Troubleshooting

Further reading

Claude Code

Intake

Log sources

​About

​Prerequisites

​Set up in the dashboard

​Store the intake URL with your agent’s configuration

​Pass the telemetry variables in your agent options

​Redeploy your agent

​Set up with the CLI

​Verify

​Troubleshooting

​Further reading

Claude Code

Intake

Log sources

About

Prerequisites

Set up in the dashboard

Store the intake URL with your agent’s configuration

Pass the telemetry variables in your agent options

Redeploy your agent

Set up with the CLI

Verify

Troubleshooting

Further reading