claude-api

Install the anthropic package with pip, create an Anthropic client that reads ANTHROPIC_API_KEY from the environment, and call client.messages.create with a model ID, max_tokens, and a messages array. The response text is available at message.content[0].text.

Define tools with a name, description, and JSON input_schema, then pass them to messages.create. When the response contains a tool_use block, execute the function with block.input and send the result back in a tool_result message to continue the conversation.

Use Sonnet 4 (claude-sonnet-4-0) for most development tasks, Opus 4.1 for complex reasoning and architecture work, and Haiku 3.5 for fast, cost-sensitive, high-volume tasks. For production, prefer pinned snapshot IDs over aliases.

Yes, both SDKs support streaming. In Python use client.messages.stream as a context manager and iterate stream.text_stream; in TypeScript iterate the stream events and handle content_block_delta events with text_delta payloads.

Use prompt caching for repeated system prompts (up to 90% savings on cached tokens), the Batches API for non-urgent bulk processing (50% off), Haiku for simple tasks (~75% cheaper than Sonnet), and shorter max_tokens when output length is predictable.

RateLimitError occurs when request volume exceeds your account limits. Catch it specifically, back off (for example sleep 60 seconds), and retry; handle APIConnectionError separately with exponential backoff for network issues.