claude-proxy (Go)

A generic API proxy that converts Anthropic/Claude API requests to OpenAI Chat Completions format and forwards to any OpenAI-compatible upstream service.

Expose POST /v1/messages (Anthropic/Claude style), convert to OpenAI Chat Completions, and proxy to upstream (configured via config.json).

Config

Edit config.json:

• ak optional: API key for inbound authentication. If set, clients must send Authorization: Bearer <ak> or x-api-key: <ak> header.
• port optional: server port (default 8888)
• upstream_timeout_seconds optional: upstream request timeout in seconds (default 300)
• log_body_max_chars optional: maximum characters to log for request/response bodies (default 4096, set to 0 to disable)
• log_stream_text_preview_chars optional: maximum characters to log for streaming response preview (default 256, set to 0 to disable)
• providers required: array of upstream service providers
  • base_url required: base URL of upstream service (e.g., https://api.example.com)
    • The completions endpoint will be constructed as {base_url}/v1/chat/completions
  • api_key required: used for upstream auth, sent as Authorization: Bearer ...
  • models required: array of model objects to expose via /v1/models endpoint
    • id required: model identifier used by clients (e.g., glm)
    • display_name optional: display name (defaults to id if not provided)
    • remote_id optional: model ID sent to upstream service (defaults to id if not provided)
• tavily_key optional: Tavily API key for web tools support.
• tavily_url optional: Tavily base URL (defaults to https://api.tavily.com).

Example:

{
  "ak": "your-proxy-api-key",
  "port": 8888,
  "upstream_timeout_seconds": 300,
  "log_body_max_chars": 4096,
  "log_stream_text_preview_chars": 256,
  "tavily_key": "tvly-...",
  "providers": [
    {
      "base_url": "https://api.example.com",
      "api_key": "your-upstream-api-key",
      "models": [
        {
          "id": "glm",
          "display_name": "glm4.7",
          "remote_id": "deepseek-chat"
        },
        {
          "id": "minimax",
          "display_name": "minimax2.1",
          "remote_id": "mimo-v2-flash"
        }
      ]
    }
  ]
}

Do not commit your real ak or api_key values.

Env

Environment variables can override or complement config.json:

• CONFIG_PATH: path to config file (default config.json)
• TAVILY_API_KEY: overrides tavily_key in config
• TAVILY_BASE_URL: overrides tavily_url in config
• TAVILY_PROXY_ADDRESS or LOCAL_PROXY_ADDRESS: proxy for Tavily requests
• TAVILY_MAX_RESULTS: max search results (default 5, max 20)
• TAVILY_SEARCH_DEPTH: basic or advanced (default basic)
• TAVILY_TOPIC: general, news, or finance (default general)
• TAVILY_FETCH_MAX_CHARS: max characters for web_fetch (default 50000, min 1000)

Run

go run .

Or build and run:

go build -o claude-proxy
./claude-proxy

CLAUDE CODE

use glm model:

export ANTHROPIC_BASE_URL=http://localhost:8888
export ANTHROPIC_AUTH_TOKEN=your-proxy-api-key
export ANTHROPIC_DEFAULT_HAIKU_MODEL=glm
export ANTHROPIC_DEFAULT_SONNET_MODEL=glm
export ANTHROPIC_DEFAULT_OPUS_MODEL=glm

claude

API

GET /v1/models

Returns the list of available models configured in config.json.

• Inbound auth:
  • If ak is set in config, you must send Authorization: Bearer <ak> (or x-api-key: <ak>).

Response format (Anthropic style):

{
  "object": "list",
  "data": [
    {
      "id": "glm",
      "object": "model",
      "created": 1234567890,
      "display_name": "glm4.7"
    }
  ]
}

POST /v1/messages

Sends a message to the upstream service.

• Inbound auth:
  • If ak is set in config, you must send Authorization: Bearer <ak> (or x-api-key: <ak>).
• Upstream auth:
  • Always sends Authorization: Bearer <api_key> to upstream.
• Web Tools:
  • Supports web_search and web_fetch if tavily_key or TAVILY_API_KEY is provided.

Example (non-stream):

curl -sS http://127.0.0.1:8888/v1/messages \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer your-proxy-api-key' \
  -d '{
    "model": "glm",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "hello"}]
  }'

GET /status

Health check endpoint.

Response:

{
  "message": "claude-proxy",
  "health": "ok"
}

Build

This project uses only Go stdlib (no external deps).

Linux (amd64):

GOOS=linux GOARCH=amd64 go build -trimpath -ldflags "-s -w" -o dist/claude-proxy_linux_amd64 .

Windows (amd64):

GOOS=windows GOARCH=amd64 go build -trimpath -ldflags "-s -w" -o dist/claude-proxy_windows_amd64.exe .

Notes / Limitations

• Multiple providers can be configured, each with their own base_url, api_key, and models.
• Web tools (web_search, web_fetch) are powered by Tavily and require an API key.
• Streaming conversion supports delta.content text and delta.tool_calls tool-use blocks.
• Logs show forwarded request bodies; keep log_body_max_chars small.
• The /v1/models endpoint returns a static list from config.