From Anthropic

If you're on the Anthropic SDK directly, you can keep using Claude through CleverRouter — but the route is AWS Bedrock in eu-central-1 rather than Anthropic's own US endpoints, and the wire format becomes OpenAI-compatible.

What's different

The gateway handles the format translation. You write OpenAI-shape requests; we send Anthropic-shape requests to Bedrock and translate the response back.

Minimal diff

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

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

const res = await client.messages.create({
  model: 'claude-3-5-sonnet-20241022',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Hello.' }],
});
console.log(res.content[0]);

import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.CLEVERROUTER_API_KEY!,
  baseURL: 'https://api.cleverouter.eu/v1',
});

const res = await client.chat.completions.create({
  model: 'anthropic/claude-3-5-sonnet:eu',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Hello.' }],
});
console.log(res.choices[0]?.message.content);

Tool calls

Anthropic's tool calls become OpenAI tool calls on the wire:

const res = await cr.chat({
  model: 'anthropic/claude-3-5-sonnet:eu',
  messages: [{ role: 'user', content: 'Weather in Paris?' }],
  tools: [
    {
      type: 'function',
      function: {
        name: 'get_weather',
        description: 'Current weather for a city.',
        parameters: {
          type: 'object',
          properties: { city: { type: 'string' } },
          required: ['city'],
        },
      },
    },
  ],
});

The gateway maps tools[] → Bedrock Converse toolConfig and flattens tool_use results back to OpenAI tool_calls. Your loop stays standard OpenAI — see Function calling.

Reasoning (extended thinking)

Anthropic's thinking.budget_tokens parameter maps to the OpenAI reasoning_effort knob:

See Reasoning.

What you give up

• Claude-3.5-Sonnet-Computer-Use — not available on Bedrock EU at time of writing.
• System-prompt caching. Anthropic's prompt-caching beta is Anthropic-direct-only; Bedrock doesn't expose it the same way.

What you keep

• The same Claude weights and behaviour.
• Streaming with SSE.
• Tool use.
• Vision.
• Long context (200k tokens, model-dependent).

Related

• Function calling
• Vision
• Why EU-native — the data-residency reason this is worth doing.