Skip to main content
Wrap your Anthropic client instance with wrapAnthropic(). Every messages.create call is traced automatically, including tool use and streaming responses.

​
Installation

npm install @zespan/sdk @anthropic-ai/sdk

​
Setup

import Anthropic from "@anthropic-ai/sdk";
import { zespan } from "@zespan/sdk";

zespan.init({ apiKey: process.env.ZESPAN_API_KEY! });

const anthropic = zespan.wrapAnthropic(new Anthropic());

​
Example

const message = await anthropic.messages.create({
  model: "claude-sonnet-4-6",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Explain observability in one paragraph." }],
});

​
What gets captured

FieldDetails
Modelclaude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5, etc.
Input tokensFrom usage.input_tokens
Output tokensFrom usage.output_tokens
CostCalculated from token counts and Anthropic pricing
LatencyTotal time from request to first token (streaming) or full response
Stop reasonend_turn, max_tokens, tool_use, stop_sequence

​
Tool use

Tool calls and tool results are captured as child spans under the message span. 
const message = await anthropic.messages.create({
  model: "claude-sonnet-4-6",
  max_tokens: 1024,
  tools: [
    {
      name: "get_weather",
      description: "Get current weather",
      input_schema: {
        type: "object",
        properties: { location: { type: "string" } },
        required: ["location"],
      },
    },
  ],
  messages: [{ role: "user", content: "What's the weather in Paris?" }],
});

​
Streaming

Streaming responses are traced end-to-end. Token counts accumulate from message_delta events. 
const stream = anthropic.messages.stream({
  model: "claude-haiku-4-5",
  max_tokens: 512,
  messages: [{ role: "user", content: "Count to ten." }],
});

for await (const chunk of stream) {
  if (chunk.type === "content_block_delta") {
    process.stdout.write(chunk.delta.text ?? "");
  }
}
Prompt and completion text is stored by default with PII redaction applied before transmission. Set storePrompts: false in zespan.init() to disable prompt storage.

​
Next steps

  • Agent tracing — trace multi-turn Claude conversations as agents
  • Guardrails — enforce safety policies on Claude outputs