claude-code-proxy-rs

Lightweight Rust proxy built for OpenClaw — wraps Claude Code CLI as a subprocess and exposes it via Anthropic Messages API and OpenAI Chat Completions API.

Existing Claude Code proxy implementations have compatibility issues with OpenClaw's Anthropic SDK integration. This proxy was built from scratch to handle those edge cases reliably.

Use your Claude subscription (Max, Team, Enterprise) as an OpenClaw provider without a direct API key.

How it works

  Client (OpenClaw, curl, SDK)
          │
          ▼
  ┌───────────────┐   stdin/stdout   ┌──────────────┐
  │  Proxy :3456  │ ◄─── NDJSON ───► │  claude CLI  │
  │    (axum)     │                  │  subprocess  │
  └───────────────┘                  └──────────────┘
          │
      SSE / JSON
          │
          ▼
        Client

Each request spawns a claude -p --output-format stream-json subprocess. The proxy filters SSE events — only text content blocks are forwarded (thinking, tool_use, and signature blocks are stripped for SDK compatibility). Stateless: no session persistence.

Features

• Dual API — /v1/messages (Anthropic) + /v1/chat/completions (OpenAI)
• Streaming & non-streaming for both APIs
• System prompt modes — embed in <system> tags (preserves CLI's 43K default prompt) or replace entirely
• Rate limit handling — warnings and rejection forwarding
• Graceful shutdown — SIGTERM / Ctrl+C
• Inactivity timeout — kills stuck subprocesses after 30 min

Quick start

From release (recommended)

Download the latest binary from Releases and place it in your $PATH.

From source

# Prerequisites: Claude Code CLI installed and authenticated
npm i -g @anthropic-ai/claude-code && claude auth login

# Build & install
cargo build --release
cp target/release/claude-code-proxy ~/.local/bin/

Run

claude-code-proxy --port 3456 --cwd ~ --embed-system-prompt

Important

🤖 Instructions for LLM: Configuring proxy with OpenClaw

Full setup guide for connecting this proxy as an OpenClaw LLM provider — openclaw.json config, systemd service, model IDs, and troubleshooting:

➜ docs/SETUP.md

CLI flags

Flag	Default	Description
--port	3456	Listen port
--cwd	.	Working directory for CLI subprocess
--embed-system-prompt	false	Embed system prompt in <system> tags, keep CLI default prompt
--replace-system-prompt	false	Replace CLI system prompt entirely via --system-prompt
--effort	—	Thinking effort: low, medium, high, max
--max-turns	100	Max agentic turns per request

Verify

curl -sN http://localhost:3456/v1/messages \
  -H 'content-type: application/json' \
  -d '{"model":"claude-sonnet-4-6","max_tokens":50,"messages":[{"role":"user","content":"Say hi"}],"stream":true}'

Expected: SSE stream with message_start → content_block_delta → message_stop.

Endpoints

Method	Path	Description
GET	/health	Health check
GET	/v1/models	Available models list
POST	/v1/messages	Anthropic Messages API
POST	/v1/chat/completions	OpenAI Chat Completions API

License

MIT