Claude Messages API

The Messages API is Anthropic’s core endpoint for talking to Claude models. You send a list of alternating user and assistant messages plus a top-level system prompt, and the API returns the model’s reply as a list of content blocks — text, tool-use requests, or thinking. Unlike OpenAI’s stateless-or-stateful hybrid, the Messages API is deliberately simple: it holds no conversation state, so every request carries the full history you want the model to see.

Knowing this API well matters because Claude models are a staple in production AI systems, and because the Messages API’s conventions — explicit system prompt, content blocks instead of plain strings, a required max_tokens — differ just enough from OpenAI’s that copying habits across providers causes real bugs. The content-block design is also what makes multimodal input (images, documents) and tool use feel uniform: everything is a typed block in the same array.

In practice you call client.messages.create() with model, max_tokens, system, and messages, then read response.content and check stop_reason to see whether the model finished naturally, hit the token cap, or wants to call a tool. You manage history yourself by appending each assistant reply and the next user turn to the array, and you enable streaming or prompt caching with the same request shape.