claude-api

# Claude API (Anthropic Messages API)

**Status**: Production Ready | **SDK**: @anthropic-ai/sdk@0.70.1

---

## Quick Start (5 Minutes)

### Node.js

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

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

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

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

### Cloudflare Workers

```typescript
const response = await fetch('https://api.anthropic.com/v1/messages', {
  method: 'POST',
  headers: {
    'x-api-key': env.ANTHROPIC_API_KEY,
    'anthropic-version': '2023-06-01',
    'content-type': 'application/json',
  },
  body: JSON.stringify({
    model: 'claude-sonnet-4-5-20250929',
    max_tokens: 1024,
    messages: [{ role: 'user', content: 'Hello!' }],
  }),
});

const data = await response.json();
console.log(data.content[0].text);
```

**Load `references/setup-guide.md` for complete setup with streaming, caching, and tools.**

---

## Critical Rules

### Always Do ✅

1. **Use environment variables** for API keys (NEVER hardcode)
2. **Set max_tokens** explicitly (required parameter)
3. **Pin model version** (`claude-sonnet-4-5-20250929`, not `claude-3-5-sonnet-latest`)
4. **Enable prompt caching** for repeated content (90% cost savings)
5. **Stream long responses** (`stream: true`) for better UX
6. **Handle errors** - Implement retry logic for 429, 529 errors
7. **Validate inputs** - Sanitize user messages before sending
8. **Monitor costs** - Track token usage
9. **Set timeouts** - Prevent hanging requests
10. **Use tool use properly** - Return tool_result in follow-up message

### Never Do ❌

1. **Never expose API key** in client-side code
2. **Never skip max_tokens** - API will error without it
3. **Never ignore stop_reason** - Check for `tool_use`, `end_turn`, `max_tokens`
4. **Never assume single content block** - `content` is an array
5. **Never use outdated models** - Pin to specific version
6. **Never skip error handling** - API calls can fail
7. **Never mix message roles** - Alternate user/assistant correctly
8. **Never ignore rate limits** - Implement exponential backoff
9. **Never store API keys** in logs or databases
10. **Never skip input validation** - Prevent injection attacks

---

## Top 3 Errors (Prevent 80% of Issues)

### Error #1: Rate Limit 429

**Symptom**: `429 Too Many Requests: Number of request tokens has exceeded your per-minute rate limit`

**Solution**: Implement exponential backoff with retry-after header

```typescript
async function handleRateLimit(requestFn, maxRetries = 3) {
  for (let attempt = 0; attempt < maxRetries; attempt++) {
    try {
      return await requestFn();
    } catch (error) {
      if (error.status === 429) {
        const retryAfter = error.response?.headers?.['retry-after'];
        const delay = retryAfter ? parseInt(retryAfter) * 1000 : 1000 * Math.pow(2, attempt);
        await new Promise(resolve => setTimeout(resolve, delay));
      } else {
        throw error;
      }
    }
  }
}
```

**Prevention**: Monitor rate limit headers, upgrade tier, implement backoff

---

### Error #2: Prompt Caching Not Activating

**Symptom**: High costs despite `cache_control` blocks, `cache_read_input_tokens: 0`

**Solution**: Place `cache_control` on LAST block with >= 1024 tokens

```typescript
// ❌ Wrong - cache_control not at end
{
  type: 'text',
  text: DOCUMENT,
  cache_control: { type: 'ephemeral' },  // Wrong position
},
{
  type: 'text',
  text: 'Additional text',
}

// ✅ Correct - cache_control at end
{
  type: 'text',
  text: DOCUMENT + '\n\nAdditional text',
  cache_control: { type: 'ephemeral' },  // Correct position
}
```

**Prevention**: Ensure content >= 1024 tokens, keep cached content identical, monitor usage

**Load `references/prompt-caching-guide.md` for complete caching strategy.**

---

### Error #3: Tool Use Response Format Errors

**Symptom**: `invalid_request_error: tools[0].input_schema is invalid`

**Solution**: Valid tool schema with proper JSON Schema

```typescript
// ✅ Valid tool schema
{
  name: 'get_weather',
  description: 'Get current weather',
  input_schema: {
    type: 'object',           // Must be 'object'
    properties: {
      location: {
        type: 'string',       // Valid JSON Schema types
        description: 'City'   // Optional but recommended
      }
    },
    required: ['location']    // List required fields
  }
}

// ✅ Valid tool result
{
  type: 'tool_result',
  tool_use_id: block.id,      // Must match tool_use id
  content: JSON.stringify(result)  // Convert to string
}
```

**Prevention**: Validate schemas, match tool_use_id exactly, stringify results

**Load `references/tool-use-patterns.md` + `references/top-errors.md` for all 12 errors.**

---

## Common Use Cases (Quick Patterns)

### Streaming Responses

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

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

**Load**: `templates/streaming-chat.ts`

---

### Prompt Caching (90% Cost Savings)

```typescript
const message = await client.messages.create({
  model: 'claude-sonnet-4-5-20250929',
  max_tokens: 1024,
  system: [
    {
      type: 'text',
      text: 'Long system prompt...',
      cache_control: { type: 'ephemeral' },
    },
  ],
  messages: [{ role: 'user', content: 'Question?' }],
});
```

**Cache lasts 5 minutes, 90% savings on cached tokens**

**Load**: `references/prompt-caching-guide.md` + `templates/prompt-caching.ts`

---

### Tool Use (Function Calling)

```typescript
const message = await client.messages.create({
  model: 'claude-sonnet-4-5-20250929',
  max_tokens: 1