Anthropic Messages

When ENABLE_ANTHROPIC_INGRESS=true, Aurora exposes Anthropic's native wire format at /v1/messages and /v1/messages/count_tokens. Point the official Anthropic SDK at Aurora and your existing Claude code works without translation.

Activate

ENABLE_ANTHROPIC_INGRESS=true
ANTHROPIC_API_KEY=sk-ant-...

Restart Aurora. The ingress is now live at http://your-aurora-host/v1/messages.

POST /v1/messages

POST /v1/messages
Authorization: Bearer <master or managed key>
anthropic-version: 2023-06-01
Content-Type: application/json
 
{
  "model": "anthropic/claude-sonnet-4-5",
  "max_tokens": 1024,
  "messages": [
    {"role": "user", "content": "Hello from Aurora"}
  ]
}

The request body is the standard Anthropic Messages API body. Aurora forwards the request to the configured Anthropic provider and returns the standard Anthropic response, including:

• content blocks: text, image, tool_use, tool_result, thinking (extended thinking)
• stop_reason and stop_sequence
• usage.input_tokens, usage.output_tokens, usage.cache_creation_input_tokens, usage.cache_read_input_tokens

Streaming is supported ("stream": true). Aurora returns the standard Anthropic SSE event sequence.

Required headers

• Authorization: Bearer <key> — your Aurora master or managed key
• anthropic-version: 2023-06-01 — Aurora requires this header
• Content-Type: application/json

POST /v1/messages/count_tokens

POST /v1/messages/count_tokens
Authorization: Bearer <master or managed key>
anthropic-version: 2023-06-01
Content-Type: application/json
 
{
  "model": "anthropic/claude-sonnet-4-5",
  "messages": [
    {"role": "user", "content": "Hello from Aurora"}
  ]
}

Response:

{
  "input_tokens": 12
}

Useful for pre-flight cost estimation without making a full request.

Using the official Anthropic SDK

from anthropic import Anthropic
 
client = Anthropic(
    base_url="http://localhost:8080",   # /v1/messages is appended automatically
    api_key="sk-aurora-...",
)
 
msg = client.messages.create(
    model="anthropic/claude-sonnet-4-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello from Aurora"}],
)
print(msg.content[0].text)

import Anthropic from "@anthropic-ai/sdk";
 
const client = new Anthropic({
  baseURL: "http://localhost:8080",
  apiKey: "sk-aurora-...",
});
 
const msg = await client.messages.create({
  model: "anthropic/claude-sonnet-4-5",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello from Aurora" }],
});
console.log(msg.content[0].text);

Tool use and extended thinking

Both are passed through with no translation:

• Tool use — tools, tool_choice, tool_use and tool_result blocks all work end-to-end.
• Extended thinking — set "thinking": {"type": "enabled", "budget_tokens": 4096} and Aurora forwards it. Thinking blocks appear in the response content as type: "thinking" and are preserved across multi-turn conversations.
• Prompt caching — cache_control: {type: "ephemeral"} on messages or system content is forwarded. Aurora reports cache_creation_input_tokens and cache_read_input_tokens in the response usage and in the dashboard.

System prompt

The Anthropic Messages API takes a top-level system field. Aurora's guardrail system prompt transforms target this exact field:

• inject — appends to the system prompt
• override — replaces the system prompt
• decorate — wraps the system prompt with prefix and suffix

Gotchas

• Anthropic-version header — required. Aurora rejects requests without it.
• Model string format — provider/model for Aurora, e.g. anthropic/claude-sonnet-4-5. If you have multiple Anthropic providers (anthropic-prod, anthropic-staging), use those names.
• Max tokens — Claude requires max_tokens. Aurora injects a default if you forget.
• Stop reason — Aurora preserves the upstream stop_reason. The OpenAI chat-completions translation has a slightly different shape; use the Anthropic SDK if you need the canonical Claude behavior.
• No passthrough needed — unlike /v1/audio/* or fine-tune endpoints, /v1/messages is fully implemented in Aurora. You do not need /p/anthropic/v1/messages.

Related

• Anthropic provider guide
• Anthropic SDK integration
• Chat Completions — OpenAI-format translation path
• Guardrails — system prompt transforms