Anthropic-compatible API

CodeGate exposes Anthropic's Messages protocol natively at /v1/messages. Request and response shapes match the Anthropic reference, so any client that speaks the Messages format (including Claude Code and the official Anthropic SDKs) will work by pointing at a different base URL.

• Base URL: https://codegate.dev
• Endpoint: POST /v1/messages
• Auth: x-api-key: sk-... plus anthropic-version: 2023-06-01
• Content-Type: application/json

POST /v1/messages

Minimal request:

curl https://codegate.dev/v1/messages \
  -H "x-api-key: sk-..." \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-opus-4-8",
    "max_tokens": 1024,
    "messages": [
      {"role": "user", "content": "Say hello."}
    ]
  }'

Response:

{
  "id": "msg_01ABC...",
  "type": "message",
  "role": "assistant",
  "model": "claude-opus-4-8",
  "content": [
    {"type": "text", "text": "Hello."}
  ],
  "stop_reason": "end_turn",
  "usage": {
    "input_tokens": 8,
    "output_tokens": 2
  }
}

Required headers

Header	Value	Purpose
x-api-key	sk-...	Your CodeGate key
anthropic-version	2023-06-01	Protocol version pinning
content-type	application/json	Request body format

If you forget anthropic-version, some SDKs will still work but the upstream may reject the request. Always send it.

Request shape

Key fields:

• model, a Claude model ID. See Models and Pricing.
• messages, an array of {role, content} pairs. role is user or assistant. content is a string or a rich content array.
• max_tokens, the maximum number of tokens in the response. Required.
• system, optional. A system prompt string or content array.
• temperature, top_p, top_k, optional sampling controls.
• stop_sequences, optional array of strings to stop generation on.
• stream, set to true to receive Server-Sent Events.
• tools, optional array of tool definitions for tool use.

Streaming (SSE)

curl -N https://codegate.dev/v1/messages \
  -H "x-api-key: sk-..." \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-opus-4-8",
    "max_tokens": 512,
    "stream": true,
    "messages": [{"role": "user", "content": "Count to five slowly."}]
  }'

The stream emits Anthropic's native SSE event types in order:

• message_start, an empty message envelope with usage counters.
• content_block_start, a new content block is beginning.
• content_block_delta, a chunk of text or partial tool input.
• content_block_stop, the block is finished.
• message_delta, updates to stop_reason, usage, and other message-level fields.
• message_stop, end of stream.

Any client that already parses this stream format works unchanged.

Tool use

Tool use is passed through end-to-end. Define tools as usual:

{
  "model": "claude-opus-4-8",
  "max_tokens": 1024,
  "tools": [
    {
      "name": "get_weather",
      "description": "Get the current weather for a city.",
      "input_schema": {
        "type": "object",
        "properties": {"city": {"type": "string"}},
        "required": ["city"]
      }
    }
  ],
  "messages": [
    {"role": "user", "content": "What is the weather in Amsterdam?"}
  ]
}

The model returns a tool_use content block. Send the tool result back in the next turn as a tool_result block. Full round-trip semantics are identical to Anthropic's direct API.

Errors

Errors are returned in Anthropic's shape:

{
  "type": "error",
  "error": {
    "type": "invalid_request_error",
    "message": "messages: at least one message is required"
  }
}

See the API overview for a status code cheat sheet.

Notes

• Use the bare host https://codegate.dev. There is no separate api. subdomain.
• CodeGate does not modify the request body before forwarding to the upstream provider (beyond routing and auth). What you send is what Claude sees.
• Response shape matches Anthropic verbatim, including stop_reason values and streaming event types.

Next

• Claude Code walkthrough for the canonical client.
• Models and Pricing for Claude per-token cost.