Anthropic Claude Integration

Anthropic Claude Integration

Setup

import Anthropic from '@anthropic-ai/sdk';
const client = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });

Basic Message

const response = await client.messages.create({
  model: 'claude-opus-4-5',
  max_tokens: 4096,
  system: 'You are Julia, a helpful agent.',
  messages: [{ role: 'user', content: 'What is 2+2?' }]
});
const text = response.content.filter(b => b.type === 'text').map(b => b.text).join('');

Multimodal (Vision)

messages: [{
  role: 'user',
  content: [
    { type: 'image', source: { type: 'base64', media_type: 'image/png', data: base64str } },
    { type: 'text', text: 'What is in this screenshot?' }
  ]
}]

Tool Use

const tools = [{
  name: 'get_weather',
  description: 'Get weather for a city',
  input_schema: { type: 'object', properties: { city: { type: 'string' } }, required: ['city'] }
}];
const response = await client.messages.create({ model, max_tokens, messages, tools });
// Check response.stop_reason === 'tool_use'
const toolUse = response.content.find(b => b.type === 'tool_use');

Streaming

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

Error Handling

• 401 → Invalid API key
• 429 → Rate limit — exponential backoff
• 529 → Overloaded — retry after delay
• Always wrap in try/catch; log e.status and e.message

Cowork-MCP Pattern

• Model: claude-opus-4-5 (default, override via env CLAUDE_MODEL)
• CHARACTER_LIMIT: 25,000 chars on output
• Stateless: new server + transport per /mcp request