Anthropic Python SDK

Why it's ranked here

A strong default for Python teams using Anthropic directly or through supported cloud providers. The package combines broad API coverage with strict typing, explicit error classes, streaming variants, and optional integrations without forcing every dependency into the base installation.

What's good

The public surface covers synchronous and asynchronous work consistently. Generated data models expose detailed content, event, capability, citation, tool, and error shapes. Separate clients support AWS, Google Cloud, Vertex, Foundry, and Bedrock. Strict Pyright settings, extensive mypy checks, warning failures, and parallel tests indicate serious maintenance discipline.

Tradeoffs

Python 3.8 and older are unsupported. Several integrations require separate optional dependency groups, and MCP support requires Python 3.10 or newer. Much of the public model surface is generated from an OpenAPI specification, so users face a large vocabulary of closely related stable and beta types.

How to use it well

Choose it for typed Python services, scripts, or workers that call Claude and need synchronous or asynchronous execution. Add only the cloud, webhook, alternate transport, or MCP extras your deployment needs. It does not serve Python 3.8 environments or replace the remote API and credentials it connects to.

Observed

License

MIT License

Primary language

Python

Install surface

Published Python package installed with pip install anthropic

Interface

Typed Python library for the Claude API with synchronous and asynchronous clients

Python support

Requires Python 3.9 or newer; MCP extra requires Python 3.10 or newer

Platform support

Classified as OS independent, with POSIX, macOS, Linux, and Windows support

Packaging

Hatchling build backend with source package under src/anthropic

Structure

Core API exports and type modules are generated from an OpenAPI specification by Stainless