goderash-adapter-anthropic

Goderash adapter for the Anthropic Python SDK — Messages + tool use.

Drop-in adapter that audits every messages.create call to Claude via Goderash — tamper-evident, hash-chained, regulator-ready. Mirrors the TypeScript adapter @goderash/adapter-claude-sdk.

Install

pip install goderash-adapter-anthropic

Quickstart

import os
import anthropic
from goderash_sdk import GoderashClient
from goderash_adapter_anthropic import wrap_anthropic, audit_messages_response

goderash = GoderashClient(
    api_key=os.environ["GODERASH_API_KEY"],
    tenant="acme",
    agent_id="claude-agent-v1",
)
ctx = goderash.new_context()

# Option 1 — wrap the client; every messages.create is audited automatically.
client = wrap_anthropic(
    client=anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"]),
    goderash=goderash,
    context=ctx,
)

response = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=1024,
    messages=[{"role": "user", "content": "hello"}],
)

# Option 2 — audit a response you already have.
audit_messages_response(goderash=goderash, context=ctx, response=response)

Events emitted per messages.create

You are responsible for emitting the matching tool.completed / tool.failed after your code runs each tool. The wrap_tool decorator in goderash-sdk does this for you automatically.

Streaming

Streaming is supported via client.messages.stream(...). The adapter aggregates token usage and emits llm.call.completed when the stream closes.

Compatibility

• anthropic >= 0.40.0
• Python >= 3.10
• Works with any Claude model (claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5, etc.)

Related packages

• goderash-sdk — core SDK + runtime guards
• goderash-adapter-openai — OpenAI Chat / Responses / Assistants
• goderash-adapter-langgraph — LangGraph callback

License

Apache-2.0