Basic Prompt Structure

The Messages API

Anthropic offers two APIs: the legacy Text Completions API and the current Messages API. This tutorial uses exclusively the Messages API.

At minimum, a call to Claude requires the following parameters:

• model — the API model name of the model you intend to call
• max_tokens — the maximum number of tokens to generate before stopping. Note that Claude may stop before reaching this maximum. This is a hard stop, meaning it may cause Claude to stop mid-word or mid-sentence.
• messages — an array of input messages. Claude is trained to operate on alternating user and assistant conversational turns. Each input message must be an object with a role and content.

There are also optional parameters:

• system — the system prompt (more below)
• temperature — the degree of variability in Claude's response. For these lessons we use temperature: 0.

Message Structure & Rules

Every message needs a role and content field:

[
  { "role": "user",      "content": "Hi Claude!" },
  { "role": "assistant", "content": "Hello! How can I help?" },
  { "role": "user",      "content": "What's the capital of France?" }
]

Two rules the API enforces:

1. Messages must alternate between user and assistant roles
2. The first message must always be a user turn

Violating either rule returns an API error.

System Prompts

A system prompt lives in a separate system parameter — not inside the messages array. It lets you provide context, instructions, and guidelines to Claude before the conversation begins.

SYSTEM: "You are a helpful assistant who only responds in rhymes."
USER:   "What's the weather like today?"
CLAUDE: "The weather today is bright and clear,
         with sunshine and warmth throughout the year!"

A well-written system prompt can dramatically improve Claude's performance — it increases Claude's ability to follow rules and stay consistent throughout a conversation.

Key Insight

Think of Claude as a capable new colleague who has zero context beyond what you explicitly tell them. Nothing is assumed. Everything must be spelled out — in either the system prompt, the user message, or both.