English | 中文

🍤 cc2api

A simple gateway that exposes the Claude Code CLI as an Anthropic-compatible API.

cc2api accepts incoming /v1/messages requests (Anthropic Messages API format), extracts the system prompt and messages, invokes claude -p via bash, and returns the response in standard Anthropic format — with full streaming (SSE) support.

Why this exists

Starting from April 4, 2026, Claude officially banned third-party OAuth calls to cc, meaning that using OAuth generated key + forged HTTP headers is no longer valid. Only requests issued through legitimate claude code are allowed.

While reading claude --help, I discovered that the claude -p/--print parameter allows for quick Q&A without entering cc's interactive UI.

So I thought, could I utilize claude -p to wrap it as an Anthropic API? Thus, this project was born.

This repository is designed for developers who want to bridge existing Anthropic-style clients with the Claude Code command-line interface. The service preserves the request shape expected by Anthropic-compatible tooling while delegating actual text generation to claude.

Key behavior

• Accepts Anthropic Messages API requests (/v1/messages)
• Supports streaming (SSE) and non-streaming responses
• Extracts system prompt and conversation messages
• Runs claude via subprocess with controlled environment variables
• Uses default system prompt: You are a helpful assistant. when none is provided
• Optional API key authentication via apikeys.txt (hot-reloaded on every request)

Supported endpoints

POST /v1/messages (Primary)

Accepts a request body in Anthropic Messages API format. Supports both streaming and non-streaming responses.

Non-streaming example:

{
  "model": "claude-sonnet-4-6",
  "max_tokens": 1024,
  "system": "You are a helpful assistant.",
  "messages": [
    {"role": "user", "content": "Translate this into bash: list all Python files recursively."}
  ],
  "effort": "high"
}

The effort field is optional. Valid values: low, medium, high, xhigh, max.

Response:

{
  "id": "msg_...",
  "type": "message",
  "role": "assistant",
  "content": [{"type": "text", "text": "..."}],
  "model": "claude-sonnet-4-6",
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {"input_tokens": 0, "output_tokens": 0}
}

Streaming example:

{
  "model": "claude-sonnet-4-6",
  "max_tokens": 1024,
  "stream": true,
  "messages": [
    {"role": "user", "content": "Explain quantum entanglement simply."}
  ]
}

The response is an SSE stream with standard Anthropic event types (message_start, content_block_delta, message_stop, etc.).

POST /v1/chat/completions (OpenAI-compatible, for testing only)

A basic OpenAI-compatible endpoint is also provided for testing purposes. Non-streaming only.

{
  "model": "claude-code",
  "messages": [
    {"role": "user", "content": "Hello!"}
  ]
}

Note: This endpoint exists for quick compatibility testing with OpenAI-style clients. It does not support streaming.

API Key Authentication

cc2api supports optional API key whitelisting. Keys are read from apikeys.txt on every request, so changes take effect immediately — no server restart required.

Setup

Create apikeys.txt in the directory where you run cc2api, one key per line:

sk-mykey-abc123
sk-another-key-xyz
# lines starting with # are ignored

When the file exists and contains at least one key, every request to /v1/messages and /v1/chat/completions must include a valid key in the Authorization header:

Authorization: Bearer sk-mykey-abc123

An invalid or missing key returns 401 Unauthorized.

When apikeys.txt does not exist or is empty, the server runs in open mode (no authentication required) — fully backwards-compatible with existing setups.

Security note: apikeys.txt is listed in .gitignore by default. Never commit real keys to version control.

Example with auth

curl -X POST http://127.0.0.1:8080/v1/messages \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-mykey-abc123" \
  -d '{
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

QuickStart

First, you need to log in to claude code cli via OAuth, which allows subsequent claude -p to run correctly internally.

Then, on a machine/terminal where you can normally use claude -p "hello?", proceed with the installation of this project:

git clone https://github.com/maxchiron/cc2api
cd cc2api
python3 -m venv .venv
source .venv/bin/activate   # macOS/Linux
# .venv\Scripts\Activate.ps1   # Windows PowerShell
pip install -e .
cc2api

Send a non-streaming request to the Anthropic endpoint:

curl -X POST http://127.0.0.1:8080/v1/messages \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Explain the single responsibility principle."}]
  }'

Send a streaming request:

curl -X POST http://127.0.0.1:8080/v1/messages \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "stream": true,
    "messages": [{"role": "user", "content": "Explain the single responsibility principle."}]
  }'

If you have configured apikeys.txt, add -H "Authorization: Bearer <your-key>" to every request.

Runtime details

The claude invocation is executed with these enforced environment variables and CLI flags:

• CLAUDE_CODE_DISABLE_AUTO_MEMORY=1
• ENABLE_CLAUDEAI_MCP_SERVERS=false
• --tools ""
• --disable-slash-commands
• --settings '{"hooks":{},"mcpServers":{}}'
• --system-prompt set from request or fallback default
• --output-format stream-json (streaming) / json (non-streaming)

The model field is passed directly to the claude CLI via --model. If the model is claude-code, --model is omitted and the CLI uses its default.

The optional effort field maps to --effort (values: low, medium, high, xhigh, max). Omitting it lets the CLI use its default effort level.

TODO

• support tool-use and more official functions. (Now not support the official format of tool-use request, but prompt-based tool-use should be working)

License

This repository is provided as-is for integration and prototyping purposes.