claude-code-proxy

Claude Code, powered by OpenAI Codex, Kimi, Grok, OpenCode Go, or Cursor Agent.

Docs: https://claude-code-proxy.raine.dev

LLM docs: https://claude-code-proxy.raine.dev/llms.txt

Claude Code remains an excellent coding harness, with strong tools, skills, hooks, subagents, and editor integrations. claude-code-proxy keeps that client experience while translating its Anthropic API traffic for subscription-backed provider services.

One local process handles provider authentication, model-based routing, protocol translation, streaming responses, and diagnostics. The built-in monitor shows sessions, active and recent requests, errors, token usage, and throughput.

Install on macOS or Linux:

brew install raine/claude-code-proxy/claude-code-proxy

Or use the release installer:

curl -fsSL https://raw.githubusercontent.com/raine/claude-code-proxy/main/scripts/install.sh | bash

Windows and other prebuilt artifacts are available from GitHub Releases.

Sign in with the Codex CLI using a ChatGPT Plus or Pro account, not an OpenAI API account. The proxy reads the credentials the Codex CLI stores in ~/.codex/auth.json and has no Codex login of its own:

codex login

Start the proxy in one terminal:

claude-code-proxy serve

For a background service, use claude-code-proxy serve --no-monitor and attach from another terminal with claude-code-proxy monitor. Closing an attached dashboard leaves the proxy running.

Start Claude Code in another:

ANTHROPIC_BASE_URL=http://127.0.0.1:18765 \
ANTHROPIC_AUTH_TOKEN=unused \
ANTHROPIC_MODEL=gpt-6-sol[1m] \
ANTHROPIC_SMALL_FAST_MODEL=gpt-6-luna[1m] \
CLAUDE_CODE_AUTO_COMPACT_WINDOW=272000 \
CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 \
CLAUDE_CODE_DISABLE_NONSTREAMING_FALLBACK=1 \
  claude

See Getting started for the complete first session.

Optional Codex image generation and editing can reuse the same ChatGPT login:

CCP_CODEX_IMAGES_API=1 claude-code-proxy serve
curl http://127.0.0.1:18765/v1/images/generations \
  -H 'Content-Type: application/json' \
  -d '{"prompt":"A paper-cut fox","model":"gpt-image-2"}'

The opt-in Images API returns base64 image data and consumes the signed-in account's image quota. Image prompts and payloads are excluded from traffic captures. See the HTTP API for generation and edit schemas.

claude-* models and the opus, sonnet, and haiku aliases are relayed to api.anthropic.com byte for byte with Claude Code's own subscription login, so no Anthropic API key is needed. gpt-* models go to Codex through the ChatGPT login above. Both run in the same session and can be switched mid-conversation with /model; reasoning from the previous backend is carried across the switch as tagged text.

Leave ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEY unset. Claude Code forwards its subscription login only when no explicit token is present; setting either one sends that value instead and the Claude route returns 401.

ANTHROPIC_BASE_URL=http://127.0.0.1:18765 \
ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-4-8 \
ANTHROPIC_DEFAULT_SONNET_MODEL=gpt-6-astra \
  claude

/model opus then uses the Claude subscription and /model gpt-6-astra uses the ChatGPT subscription. CCP_ANTHROPIC_BASE_URL overrides the Anthropic upstream. Codex credentials are read from ~/.codex/auth.json; CODEX_HOME or CCP_CODEX_AUTH_FILE points the proxy at another file, and refreshed tokens are written back so the Codex CLI keeps working.

Claude Code turns on its /effort menu by matching the model id against its own patterns, so a gpt-* id gets no effort menu until its capabilities are declared as a custom picker entry:

export ANTHROPIC_CUSTOM_MODEL_OPTION="gpt-6-astra"
export ANTHROPIC_CUSTOM_MODEL_OPTION_SUPPORTED_CAPABILITIES="effort,xhigh_effort,max_effort"

Run claude-code-proxy models for the current catalog or claude-code-proxy models --full for every dynamic Cursor alias.

• What is claude-code-proxy?
• Choosing a provider
• Configure Claude Code
• Models and routing
• Monitor TUI
• Troubleshooting
• Command reference
• Configuration
• HTTP API
• Compatibility and limitations
• For coding agents

• aven: local-first task management for power users and agents
• claude-history: search Claude Code conversation history from the terminal
• git-surgeon: non-interactive hunk-level git staging for coding agents
• workmux: parallel coding tasks in git worktrees and tmux
• consult-llm: consult other AI models from an agent workflow

MIT