telemetry-dev-anthropic

Anthropic Claude SDK instrumentation for telemetry.dev. It wraps the official anthropic Python SDK and emits telemetry.dev generation spans through telemetry-dev.

Install

pip install telemetry-dev-anthropic

Initialize the core SDK first:

import telemetry_dev

telemetry_dev.init(
    api_key="td_live_...",
    service_name="my-service",
)

Per-client wrapping

from anthropic import Anthropic
from telemetry_dev_anthropic import wrap_anthropic

client = wrap_anthropic(Anthropic())
message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Tell me a joke about OpenTelemetry"}],
)

wrap_anthropic also supports AsyncAnthropic and the provider clients AnthropicBedrock, AnthropicBedrockMantle, AnthropicVertex, AnthropicAWS, and AnthropicFoundry, with their Async variants.

Global instrumentation

from anthropic import Anthropic
from telemetry_dev_anthropic import instrument_anthropic, uninstrument_anthropic

instrument_anthropic()
try:
    client = Anthropic()
    client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
        messages=[{"role": "user", "content": "Hello"}],
    )
finally:
    uninstrument_anthropic()

Instrumented surfaces

• client.messages.create(...)
• client.messages.create(..., stream=True)
• client.messages.stream(...) context managers, sync and async
• client.messages.parse(...), sync and async
• client.beta.messages.create(...), stream(...), and parse(...), sync and async
• client.beta.messages.tool_runner(...): each model request in the tool loop records its own generation span

The integration maps native Anthropic request and response shapes directly into telemetry.dev fields. It does not normalize messages into another schema.

Streaming

Native Anthropic stream events pass through unmodified. The span records time to first chunk on the first received event, merges usage from message_start and message_delta, aggregates text, tool-use JSON, and thinking blocks, and ends on stream exhaustion, close, context-manager exit, or error.

Reconstructed output is bounded to 48 KiB and 1,000 items. The configured max_attribute_length does not shrink these retention bounds; the core applies it to the exported attribute after the mask runs, with its ...[truncated] marker, without setting telemetry.dev.capture.truncated. A stream that ends without message_stop, encounters a mapping failure, or exceeds a capture bound sets telemetry.dev.capture.truncated. With a mask configured, incomplete output is omitted because the mask cannot inspect the complete value. When output capture is disabled, response IDs and models, finish reasons, usage, timing, and errors may still be recorded. Capture flags do not gate stop sequences, caller-supplied metadata or raw attributes, or exception messages and stack traces; redact those separately when needed. Request tools and tool_choice are part of the captured input, so capture_input=False removes them.

messages.stream() starts the span when the context manager is entered, because that is when the Anthropic SDK opens the HTTP stream.

Provider clients

Class instrumentation covers the provider clients. Their messages resources reuse the stable Messages and AsyncMessages classes. The Bedrock and Vertex beta.messages resources are separate provider classes, which are patched too: Bedrock exposes create() and Vertex exposes create() and stream(). Provider attribution comes from the client class, including subclasses: aws.bedrock for AnthropicBedrock and AnthropicBedrockMantle, and gcp.vertex_ai for AnthropicVertex. AnthropicAWS (Claude Platform on AWS) and AnthropicFoundry (Microsoft Foundry) serve the Anthropic API, so they are recorded as anthropic.

Limitations

• messages.count_tokens(), beta.messages.count_tokens(), and batches are not instrumented.
• with_raw_response snapshots bound methods; wrap or instrument clients before creating raw-response wrappers.
• Unconsumed streams end spans only on exhaustion, close, context-manager exit, or error.