Working with Anthropic Claude API

Anthropic's Claude models are known for long context windows, strong reasoning, and a focus on being helpful, harmless, and honest. This topic covers how to build with the Claude API.

Setting Up

1. Get API Access

bash

export ANTHROPIC_API_KEY="sk-ant-your-key-here"

2. Install the SDK

bash

# Python
pip install anthropic

# Node.js
npm install @anthropic-ai/sdk

Your First Claude API Call

Python

python

import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Explain recursion in simple terms."}
    ]
)

print(message.content[0].text)

Node.js

javascript

import Anthropic from '@anthropic-ai/sdk';

const anthropic = new Anthropic();

const message = await anthropic.messages.create({
    model: "claude-sonnet-4-20250514",
    max_tokens: 1024,
    messages: [
        { role: "user", content: "Explain recursion in simple terms." }
    ]
});

console.log(message.content[0].text);

Claude's Message Format

Claude uses a similar messages structure to OpenAI, with one key difference — the system prompt is a separate parameter:

python

message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    system="You are an expert Python developer. Be concise and practical.",
    messages=[
        {"role": "user", "content": "How do I handle errors in async Python?"},
        {"role": "assistant", "content": "Use try/except within async functions..."},
        {"role": "user", "content": "Show me a real-world example."}
    ]
)

Available Models

Model	Strengths	Context Window
Claude Opus 4	Most capable, complex reasoning	200K tokens
Claude Sonnet 4	Best balance of speed and quality	200K tokens
Claude Haiku 3.5	Fastest, most affordable	200K tokens

Choosing the Right Model

Haiku: Simple tasks, classification, extraction, high-volume
Sonnet: General-purpose, code generation, analysis
Opus: Complex reasoning, research, nuanced tasks

Key Parameters

python

message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=2048,               # Required: max response tokens
    system="System instructions",   # Optional: system prompt
    messages=[...],                 # Required: conversation
    temperature=0.7,                # Optional: 0-1 (default: 1)
    top_p=0.9,                      # Optional: nucleus sampling
    stop_sequences=["\n\nHuman:"],  # Optional: stop generation
)

Streaming

python

with client.messages.stream(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Write a poem about coding"}]
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

Working with Long Documents

Claude's 200K token context window (~150K words) is ideal for processing long documents:

python

# Read a long document
with open("annual_report.txt") as f:
    document = f.read()

message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=2048,
    messages=[
        {
            "role": "user",
            "content": f"""Analyze this annual report and provide:
1. Key financial highlights
2. Major risks identified
3. Strategic priorities for next year

<document>
{document}
</document>"""
        }
    ]
)

Vision (Image Understanding)

Claude can analyze images:

python

import base64

with open("screenshot.png", "rb") as f:
    image_data = base64.standard_b64encode(f.read()).decode("utf-8")

message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "image",
                    "source": {
                        "type": "base64",
                        "media_type": "image/png",
                        "data": image_data
                    }
                },
                {
                    "type": "text",
                    "text": "Describe what you see in this screenshot."
                }
            ]
        }
    ]
)

Tool Use (Function Calling)

Claude can call functions you define:

python

message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    tools=[
        {
            "name": "get_weather",
            "description": "Get current weather for a location",
            "input_schema": {
                "type": "object",
                "properties": {
                    "location": {
                        "type": "string",
                        "description": "City name, e.g., 'London'"
                    }
                },
                "required": ["location"]
            }
        }
    ],
    messages=[
        {"role": "user", "content": "What's the weather in Tokyo?"}
    ]
)

# Check if Claude wants to use a tool
for block in message.content:
    if block.type == "tool_use":
        print(f"Tool: {block.name}")
        print(f"Input: {block.input}")
        # → Tool: get_weather
        # → Input: {"location": "Tokyo"}

XML Tags — Claude's Secret Weapon

Claude responds especially well to XML-structured prompts:

python

message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=2048,
    messages=[
        {
            "role": "user",
            "content": """
<context>
You are reviewing a Node.js Express application for security vulnerabilities.
The application handles payment processing.
</context>

<code>
app.post('/pay', (req, res) => {
    const amount = req.body.amount;
    const query = `UPDATE accounts SET balance = balance - ${amount}`;
    db.execute(query);
    res.json({ success: true });
});
</code>

<task>
Identify all security vulnerabilities and provide fixed code.
</task>
"""
        }
    ]
)

Error Handling

python

import anthropic

client = anthropic.Anthropic()

try:
    message = client.messages.create(
        model="claude-sonnet-4-20250514",
        max_tokens=1024,
        messages=[{"role": "user", "content": "Hello"}]
    )
except anthropic.RateLimitError:
    print("Rate limited — implement backoff")
except anthropic.APIConnectionError:
    print("Connection failed — check network")
except anthropic.APIStatusError as e:
    print(f"API error: {e.status_code}")

Usage Tracking

python

message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}]
)

print(f"Input tokens:  {message.usage.input_tokens}")
print(f"Output tokens: {message.usage.output_tokens}")

Claude vs OpenAI: Key API Differences

Feature	Claude	OpenAI
System prompt	Separate `system` parameter	In messages array
Response format	`message.content[0].text`	`response.choices[0].message.content`
Max tokens	Required parameter	Optional
Context window	200K tokens	4K–128K
Streaming	`messages.stream()` context manager	`stream=True` parameter
XML in prompts	Highly effective	Works but less optimized

Best Practices for Claude

Use XML tags — Claude handles structured XML prompts exceptionally well
Leverage the long context — Process entire documents, codebases, or conversation histories
Be specific in system prompts — Claude follows system instructions closely
Use Haiku for simple tasks — Save costs on classification, extraction, routing
Use Sonnet for code — Best balance of quality and speed for development tasks

Next, we'll explore Google's Gemini API and then compare all three providers.