Anthropic

About the Anthropic integration

@telemetry-dev/anthropic (TypeScript) and telemetry-dev-anthropic (Python) wrap the official Anthropic SDKs and emit OpenTelemetry GenAI spans. messages.create and the messages.stream helper are traced — including stream: true responses — with model, system instructions, tools, messages, finish reason, and token usage with cache and thinking token details when Anthropic returns them.

Key features

• Full Messages coverage: Requests capture model, max tokens, sampling params, stop sequences, system instructions, tools, and messages; responses capture content blocks, finish reason, and usage.
• Streaming done right: Spans start at request time and aggregate text, tool input fragments, thinking deltas, and usage until the stream is consumed, closed, or errors.
• Bedrock and Vertex clients too: Anthropic, Bedrock, and Vertex clients are all supported, with provider recorded as anthropic, aws.bedrock, or gcp.vertex_ai.
• Cache and thinking tokens: Cache read/write and thinking token details flow through when the API returns them, so cost stays accurate.
• Per-client or global: wrapAnthropic(client) is idempotent; instrumentAnthropic() patches once at startup.

Get started

import Anthropic from "@anthropic-ai/sdk";
import { init, shutdown } from "@telemetry-dev/sdk";
import { wrapAnthropic } from "@telemetry-dev/anthropic";

init({ serviceName: "anthropic-worker" });

const anthropic = wrapAnthropic(new Anthropic());
await anthropic.messages.create({
  model: "claude-sonnet-4-6",
  max_tokens: 256,
  messages: [{ role: "user", content: "Say hello." }],
});

await shutdown();

In Python, wrap_anthropic also covers AsyncAnthropic, AnthropicBedrock, and AnthropicVertex clients:

import telemetry_dev
from anthropic import Anthropic
from telemetry_dev_anthropic import wrap_anthropic

telemetry_dev.init()

client = wrap_anthropic(Anthropic())
client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Say hello"}],
)