Anthropic (JS/TS)

Trace Anthropic Claude calls from JavaScript/TypeScript using OpenTelemetry instrumentation and Muster.

This guide demonstrates how to combine Anthropic's JavaScript/TypeScript SDK with Muster for comprehensive AI application observability. The approach uses OpenTelemetry instrumentation to automatically capture and monitor Anthropic SDK interactions.

Setup

Installation

npm install @anthropic-ai/sdk @arizeai/openinference-instrumentation-anthropic @langfuse/otel @opentelemetry/sdk-node

Environment Configuration

Set credentials for both Anthropic and Muster:

ANTHROPIC_API_KEY="sk-ant-..."

LANGFUSE_PUBLIC_KEY="pk-lf-..."
LANGFUSE_SECRET_KEY="sk-lf-..."
LANGFUSE_BASE_URL="https://app.getmuster.io"   # or your self-hosted Muster URL

Anthropic API keys are issued from the Anthropic Console. Muster API keys are issued from your project's Settings → API Keys page.

OpenTelemetry Initialization

import Anthropic from "@anthropic-ai/sdk";
import { NodeSDK } from "@opentelemetry/sdk-node";
import { LangfuseSpanProcessor } from "@langfuse/otel";
import { AnthropicInstrumentation } from "@arizeai/openinference-instrumentation-anthropic";

const instrumentation = new AnthropicInstrumentation();
instrumentation.manuallyInstrument(Anthropic);

const sdk = new NodeSDK({
  spanProcessors: [new LangfuseSpanProcessor()],
  instrumentations: [instrumentation],
});

sdk.start();

After initialization, every call you make through the Anthropic SDK is captured automatically as a span and forwarded to your Muster project.

const client = new Anthropic();

const message = await client.messages.create({
  model: "claude-opus-4-20250514",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello, Claude" }],
});

console.log(message);

Advanced Integration Features

The integration supports deeper customization through:

Context Managers — wrap instrumented code to add attributes and establish observation hierarchies.
observe wrapper — decorate existing functions for automatic span creation without code modification.
Attribute propagation — distribute userId, sessionId, metadata, tags, and version information across child observations.

See the JS/TS SDK reference for the full API.

Troubleshooting

Enable debug logging to diagnose tracing issues:

LANGFUSE_LOG_LEVEL="DEBUG"

Other things to check:

In short-lived environments, call await langfuseSpanProcessor.forceFlush() before exit so spans are not dropped.
Filter unwanted OpenTelemetry spans from unrelated libraries — they still count toward your billable units.