Skip to main content
NPM Version This module provides OpenInference instrumentation for Anthropic’s Claude Agent SDK, automatically capturing AGENT and TOOL spans that follow OpenInference semantic conventions.

​
Install

npm install @arizeai/openinference-instrumentation-claude-agent-sdk @arizeai/phoenix-otel

​
Setup

To instrument your Claude Agent SDK application, use the register function from @arizeai/phoenix-otel and set up the Claude Agent SDK instrumentation. Create the instrumentation.ts file: 
import { register } from "@arizeai/phoenix-otel";
import { ClaudeAgentSDKInstrumentation } from "@arizeai/openinference-instrumentation-claude-agent-sdk";
import * as ClaudeAgentSDK from "@anthropic-ai/claude-agent-sdk";

// Initialize Phoenix tracing
const tracerProvider = register({
  projectName: "claude-agent-sdk-app",
  // If using Phoenix Cloud:
  // url: "https://app.phoenix.arize.com/s/your-space-name",
  // apiKey: process.env.PHOENIX_API_KEY,
  // If using self-hosted Phoenix:
  // url: "http://localhost:6006",
});

// Set up Claude Agent SDK instrumentation
const instrumentation = new ClaudeAgentSDKInstrumentation();
instrumentation.manuallyInstrument(ClaudeAgentSDK);

console.log("Claude Agent SDK instrumentation registered");
The Claude Agent SDK is ESM-only, so manuallyInstrument() is used instead of enable() to ensure compatibility. You must import the SDK namespace and pass it to manuallyInstrument().

​
Usage

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

async function main() {
  for await (const message of query({
    prompt: "What is the weather in San Francisco?",
    options: {
      model: "claude-sonnet-4-5-20250514",
    },
  })) {
    if (message.type === "assistant") {
      console.log(message.content);
    }
  }
}

main();

​
Observe

With instrumentation enabled, you will see the following in Phoenix:
  • AGENT spans wrapping the full query() call, capturing the prompt and final response
  • TOOL spans for each tool invocation made by the agent during execution
These spans follow OpenInference semantic conventions, making them fully compatible with Phoenix’s trace visualization and evaluation features.

​
Privacy Configuration

You can configure the instrumentation to hide sensitive data using traceConfig: 
const instrumentation = new ClaudeAgentSDKInstrumentation({
  traceConfig: {
    hideInputs: true, // Omit input content from spans
    hideOutputs: true, // Omit output content from spans
  },
});
instrumentation.manuallyInstrument(ClaudeAgentSDK);

​
Resources