Claude SDK Integration Patterns

Production-ready patterns for integrating Claude API and TypeScript SDK into applications. Master streaming responses, tool execution, error handling, and optimization strategies for AI-powered features.

When to Use This Skill

Use this skill when you need to:

Integrate Claude API into Node.js/TypeScript applications
Implement streaming conversations with real-time responses
Build applications with Claude tool use (function calling)
Handle API errors gracefully with retry logic
Optimize token usage and manage costs
Deploy Claude-powered features to production
Build multi-turn conversations with context management
Implement message batching for high-volume processing

Core Concepts

Messages API Fundamentals

The Claude Messages API is the primary interface for conversational AI:

Key Components:

Model Selection: Choose appropriate model (Opus, Sonnet, Haiku)
Messages Array: Conversation history with user/assistant roles
Max Tokens: Control response length
System Prompts: Guide model behavior
Streaming: Real-time response generation

SDK Architecture

The TypeScript SDK provides:

Type-safe API client
Streaming helpers for real-time responses
Tool execution framework
Error handling utilities
Message batch processing
Event-driven architecture

Installation and Setup

Install the SDK

npm install @anthropic-ai/sdk

Or with yarn

yarn add @anthropic-ai/sdk

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

// Initialize client const anthropic = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY, });

Messages API Patterns

Pattern 1: Basic Message Creation

const message = await anthropic.messages.create({ model: 'claude-sonnet-4-5-20250929', max_tokens: 1024, messages: [ { role: 'user', content: 'Hello, Claude!' } ], });

console.log(message.content);

When to use:

Simple question-answer interactions
One-off API calls
Synchronous workflows

Pattern 2: Multi-Turn Conversations

const messages = [ { role: 'user', content: 'What is TypeScript?' }, { role: 'assistant', content: 'TypeScript is a typed superset of JavaScript...' }, { role: 'user', content: 'Give me an example' }, ];

const response = await anthropic.messages.create({ model: 'claude-sonnet-4-5-20250929', max_tokens: 1024, messages, });

When to use:

Chatbots and conversational UIs
Multi-step workflows
Context-dependent interactions

Pattern 3: System Prompts

const response = await anthropic.messages.create({ model: 'claude-sonnet-4-5-20250929', max_tokens: 1024, system: 'You are a helpful Python programming assistant. Provide concise, tested code examples.', messages: [ { role: 'user', content: 'How do I read a CSV file?' } ], });

When to use:

Specialized assistants
Role-playing scenarios
Consistent behavior across conversations

Streaming Patterns

Pattern 4: Basic Streaming

const stream = await anthropic.messages.create({ model: 'claude-sonnet-4-5-20250929', max_tokens: 1024, messages: [{ role: 'user', content: 'Write a story' }], stream: true, });

for await (const event of stream) { if (event.type === 'content_block_delta' && event.delta.type === 'text_delta') { process.stdout.write(event.delta.text); } }

When to use:

Real-time user interfaces
Long-form content generation
Interactive experiences

Pattern 5: Streaming with Event Handlers

const finalMessage = await stream.finalMessage();

When to use:

Real-time UIs (chatbots, live editors)
Progress indicators
Partial result processing

Pattern 6: Streaming with Abort Control

const stream = anthropic.messages.stream({ model: 'claude-sonnet-4-5-20250929', max_tokens: 1024, messages: [{ role: 'user', content: 'Long task...' }], });

// Abort after timeout setTimeout(() => stream.abort(), 5000);

try { await stream.done(); } catch (error) { if (error instanceof Anthropic.APIUserAbortError) { console.log('Stream aborted by user'); } }

When to use:

User-cancellable operations
Timeout handling
Resource management

Tool Use Patterns

Pattern 7: Tool Definition with Zod

import { betaZodTool } from '@anthropic-ai/sdk/helpers/zod'; import { z } from 'zod';

const weatherTool = betaZodTool({ name: 'get_weather', inputSchema: z.object({ location: z.string(), unit: z.enum(['celsius', 'fahrenheit']).default('fahrenheit'), }), description: 'Get current weather for a location', run: async (input) => { // Call weather API return Weather in ${input.location}: 72°F, sunny; }, });

When to use:

Type-safe tool definitions
Input validation
Clear tool contracts

Pattern 8: Tool Runner for Automatic Execution

const finalMessage = await anthropic.beta.messages.toolRunner({ model: 'claude-sonnet-4-5-20250929', max_tokens: 1000, messages: [ { role: 'user', content: 'What's the weather in San Francisco?' } ], tools: [weatherTool], });

console.log(finalMessage.content);

When to use:

Automated tool execution
AI agents with function calling
Complex multi-step workflows

Pattern 9: Streaming Tool Execution

const runner = anthropic.beta.messages.toolRunner({ model: 'claude-sonnet-4-5-20250929', max_tokens: 1000, messages: [{ role: 'user', content: 'Check weather and calculate travel time' }], tools: [weatherTool, travelTimeTool], stream: true, });

for await (const messageStream of runner) { for await (const event of messageStream) { console.log('Event:', event); } console.log('Message:', await messageStream.finalMessage()); }

When to use:

Real-time tool execution feedback
Multi-tool workflows
Interactive AI agents

Error Handling Patterns

Pattern 10: Comprehensive Error Handling

async function createMessage(prompt: string) { try { const message = await anthropic.messages.create({ model: 'claude-sonnet-4-5-20250929', max_tokens: 1024, messages: [{ role: 'user', content: prompt }], });

return message;

} catch (error) { if (error instanceof Anthropic.APIError) { console.error('API Error:', { status: error.status, name: error.name, message: error.message, headers: error.headers, });

  // Handle specific errors
  if (error.status === 429) {
    // Rate limit - implement backoff
    console.log('Rate limited, waiting...');
    await new Promise(resolve => setTimeout(resolve, 60000));
    return createMessage(prompt); // Retry
  } else if (error.status === 401) {
    throw new Error('Invalid API key');
  } else if (error.status === 400) {
    throw new Error(`Bad request: ${error.message}`);
  }
}

throw error;

} }

When to use:

Production applications
Robust error recovery
User-facing applications

Pattern 11: Exponential Backoff Retry

async function createWithRetry( params: Anthropic.MessageCreateParams, maxRetries = 3, baseDelay = 1000 ): Promise<Anthropic.Message> { for (let attempt = 0; attempt < maxRetries; attempt++) { try { return await anthropic.messages.create(params); } catch (error) { if (error instanceof Anthropic.APIError && error.status === 429) { // Rate limit - exponential backoff const delay = baseDelay * Math.pow(2, attempt); console.log(Retry attempt ${attempt + 1} after ${delay}ms); await new Promise(resolve => setTimeout(resolve, delay)); } else { throw error; } } }

throw new Error('Max retries exceeded'); }

When to use:

High-volume applications
Rate limit handling
Network reliability

Token Optimization Patterns

Pattern 12: Token Counting

import { encode } from 'gpt-tokenizer';

function estimateTokens(text: string): number { // Approximate token count (Claude uses similar tokenization to GPT) return encode(text).length; }

function optimizePrompt(userMessage: string, maxTokens: number = 1024): string { const tokens = estimateTokens(userMessage);

if (tokens > maxTokens) { // Truncate or summarize return userMessage.slice(0, maxTokens * 4); // Rough char estimate }

return userMessage; }

When to use:

Cost optimization
Token budget management
Large input handling

Pattern 13: Context Window Management

interface ConversationMessage { role: 'user' | 'assistant'; content: string; tokens?: number; }

class ConversationManager { private messages: ConversationMessage[] = []; private maxContextTokens = 100000;

addMessage(role: 'user' | 'assistant', content: string) { const tokens = estimateTokens(content); this.messages.push({ role, content, tokens }); this.pruneIfNeeded(); }

private pruneIfNeeded() { const totalTokens = this.messages.reduce((sum, msg) => sum + (msg.tokens || 0), 0);

if (totalTokens > this.maxContextTokens) {
  // Remove oldest messages (keep system prompt if present)
  this.messages = this.messages.slice(-10); // Keep last 10 messages
}

}

getMessages() { return this.messages.map(({ role, content }) => ({ role, content })); } }

When to use:

Long conversations
Context management
Cost control

Message Batching Patterns

Pattern 14: Batch Processing

const batchResult = await anthropic.messages.batches.create({ requests: [ { custom_id: 'request-1', params: { model: 'claude-sonnet-4-5-20250929', max_tokens: 1024, messages: [{ role: 'user', content: 'Summarize this article...' }], }, }, { custom_id: 'request-2', params: { model: 'claude-sonnet-4-5-20250929', max_tokens: 1024, messages: [{ role: 'user', content: 'Translate this text...' }], }, }, ], });

// Poll for results const batch = await anthropic.messages.batches.retrieve(batchResult.id);

When to use:

Bulk processing
Offline workflows
Cost optimization (lower batch pricing)

Production Patterns

Pattern 15: Request Timeout Configuration

const anthropic = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY, timeout: 60000, // 60 seconds maxRetries: 3, });

When to use:

Production deployments
Network reliability
Resource management

Pattern 16: Custom Headers and Metadata

const message = await anthropic.messages.create( { model: 'claude-sonnet-4-5-20250929', max_tokens: 1024, messages: [{ role: 'user', content: 'Hello' }], }, { headers: { 'anthropic-beta': 'max-tokens-3-5-sonnet-2024-07-15', }, } );

When to use:

Beta features
Custom tracking
A/B testing

Best Practices

API Key Management

Never hardcode API keys

// ❌ Bad const anthropic = new Anthropic({ apiKey: 'sk-ant-...' });

// ✅ Good const anthropic = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });

Use environment variables

.env file

ANTHROPIC_API_KEY=sk-ant-your-key-here

Model Selection

Choose appropriate model for task:

Opus: Most capable, complex reasoning
Sonnet: Balanced performance/speed
Haiku: Fastest, simple tasks

Use latest model versions:

model: 'claude-sonnet-4-5-20250929' // Latest

Streaming Considerations

Always use streaming for:

User-facing applications
Long-form content (>500 tokens)
Real-time interactions

Avoid streaming for:

Backend batch processing
Deterministic outputs needed upfront
Simple API integrations

Error Handling

Always catch API errors
Implement retry logic for transient failures
Log errors for debugging
Provide user-friendly error messages

Token Management

Estimate tokens before API calls
Set appropriate max_tokens
Monitor token usage
Implement context pruning for long conversations

Troubleshooting

Issue: Rate Limit Errors (429)

Solution:

Implement exponential backoff
Use message batching for bulk operations
Upgrade to higher rate limit tier

Issue: Timeout Errors

Solution:

Use streaming for long responses
Increase timeout configuration
Reduce max_tokens

Issue: Invalid API Key (401)

Solution:

Verify environment variable is set
Check API key format (starts with sk-ant- )
Ensure key hasn't been rotated

Issue: Context Length Exceeded

Solution:

Implement conversation pruning
Summarize older messages
Use appropriate context window limits

Resources

Anthropic API Documentation
TypeScript SDK GitHub
Claude API Pricing
Model Comparison

Skill Version: 1.0.0 Last Updated: October 2025 Compatibility: @anthropic-ai/sdk 0.30.0+ Research Source: Context7 - Anthropic TypeScript SDK (106 code snippets, Trust Score 8.8)

claude-sdk-integration-patterns

Safety Notice

Copy this and send it to your AI assistant to learn

Install the SDK

Or with yarn

.env file

Source Transparency

Related Skills

docker-compose-orchestration

postgresql-database-engineering

jest-react-testing

ui-design-patterns