You're already paying for Claude. Why pay again for the API?
If you have a Claude Pro ($20/mo) or Max ($100-200/mo) subscription, you already have access to Opus, Sonnet, and Haiku. But most AI-powered IDEs and agent frameworks want an API key — which means paying a second time at $15/MTok for Opus.
OCP (OpenClaw Claude Proxy) solves this. It exposes your Claude subscription as a standard OpenAI-compatible API on localhost. Any tool that speaks the OpenAI protocol can use it:
- OpenClaw — multi-agent orchestration
- OpenCode — AI coding assistant
- Cline — VS Code AI extension
- Continue.dev — open-source AI assistant
- Aider — AI pair programming
- Any tool that accepts an OpenAI base URL
One subscription. Multiple instances. All IDEs. $0 extra.
How It Works
Your IDE / Agent → OCP (localhost:3456) → claude CLI → Anthropic (via your subscription)
OCP translates OpenAI-compatible /v1/chat/completions requests into claude -p CLI calls. Anthropic sees normal Claude Code usage under your existing subscription. No separate API key needed.
# Point any OpenAI-compatible tool at OCP
export OPENAI_BASE_URL=http://127.0.0.1:3456/v1
export OPENAI_API_KEY=dummy # OCP doesn't need a real key
Setup in 60 Seconds
# Prerequisites: Node.js 18+, Claude CLI installed & authenticated
git clone https://github.com/dtzp555-max/openclaw-claude-proxy
cd openclaw-claude-proxy
node setup.mjs # auto-configures + installs as system service
That's it. The proxy runs as a daemon (launchd on macOS, systemd on Linux), auto-starts on boot, and auto-recovers from crashes. Your IDE doesn't know the difference — it just sees an OpenAI API.
Monitor Your Usage in Real Time
With a Max subscription shared across multiple IDEs and agents, you need to know where you stand. The built-in ocp CLI gives you instant visibility:
$ ocp usage
Plan Usage Limits
─────────────────────────────────────
Current session 3% used
Resets in 4h 32m (Tue, Mar 24, 10:00 PM)
Weekly (all models) 3% used
Resets in 6d 6h (Tue, Mar 31, 12:00 AM)
Extra usage off
Model Stats
Model Req OK Er AvgT MaxT AvgP MaxP
──────────────────────────────────────────────────────
haiku 1 1 0 6s 6s 0K 0K
opus 2 2 0 20s 26s 42K 43K
sonnet 2 2 0 24s 24s 41K 41K
Total 5
Proxy: up 0h 37m | 5 reqs | 0 err | 0 timeout
See your session utilization, weekly limits, per-model request stats, response times, and prompt sizes — all at a glance.
If you use OpenClaw with Telegram or Discord, these commands also work as native slash commands: /ocp usage, /ocp status, /ocp settings.
Tune Settings Without Restart
Running Opus with large prompts? Bump the timeout. Too many concurrent agents? Cap the limit. All at runtime:
$ ocp settings maxPromptChars 200000
✓ maxPromptChars = 200000
$ ocp settings tiers.opus.base 180000
✓ tiers.opus.base = 180000
Tunable parameters: timeout, firstByteTimeout, maxConcurrent, sessionTTL, maxPromptChars, and per-model timeout tiers.
Available Models
| Model ID | Maps to | Notes |
|---|---|---|
claude-opus-4-6 |
Claude Opus 4.6 | Most capable |
claude-sonnet-4-6 |
Claude Sonnet 4.6 | Best speed/quality balance |
claude-haiku-4 |
Claude Haiku 4.5 | Fastest, lightweight tasks |
All OCP Commands
ocp usage Session/weekly limits & model stats
ocp status Quick health overview
ocp health Full proxy diagnostics
ocp settings View all tunable parameters
ocp settings <k> <v> Change a setting at runtime
ocp logs [N] [level] Recent proxy logs
ocp models List available models
ocp sessions Active CLI sessions
ocp clear Clear all sessions
ocp restart Restart proxy or gateway
Multi-Instance Sharing
The real power: one proxy, multiple consumers. I run 6 OpenClaw agents + Claude Code interactive sessions, all hitting the same OCP instance on localhost:3456. The proxy handles concurrency (default: 8 parallel requests), session management, and adaptive timeouts per model.
OpenClaw Agent 1 (Sonnet) ──┐
OpenClaw Agent 2 (Opus) ───┤
OpenCode (Sonnet) ───┼──→ OCP :3456 ──→ claude CLI ──→ Your subscription
Cline (Haiku) ───┤
Aider (Sonnet) ───┘
All sharing the same Max subscription. Check ocp usage anytime to see how much headroom you have.
Built-in Safety
- Prompt truncation — auto-caps prompts at 150K chars to prevent runaway context from eating your quota
- Adaptive timeouts — Opus gets 150s, Sonnet 120s, Haiku 45s base first-byte timeout, scaling with prompt size
- Localhost only — binds to 127.0.0.1, not exposed to the network
-
Optional Bearer auth — set
PROXY_API_KEYif you want token protection
Try It
git clone https://github.com/dtzp555-max/openclaw-claude-proxy
cd openclaw-claude-proxy
node setup.mjs
ln -sf $(pwd)/ocp /usr/local/bin/ocp
ocp usage # see your subscription status
GitHub: dtzp555-max/openclaw-claude-proxy
Requirements: Node.js 18+, Claude CLI authenticated (claude login)
How much are you spending on AI API keys across different tools? I'm curious if others are looking for ways to consolidate.
Top comments (0)