ai-cli-bridge – Claude Max and ChatGPT Pro as a Team API

TL;DR

TL;DR: ai-cli-bridge wraps Claude Code CLI and OpenAI Codex CLI behind an Express API, letting teams share a single Claude Max or ChatGPT Pro subscription as a metered HTTP API with per-key auth and rate limits.

What Is ai-cli-bridge?

If you pay for Claude Max ($100–200/mo) or OpenAI Pro ($200/mo), that usage is locked to the vendor’s own CLI tools — they cannot be called from web apps, plugins, or arbitrary HTTP clients. ai-cli-bridge solves that by wrapping both CLIs behind a self-hosted REST API.

The core idea:

Approach	Auth	Billing
SDK (Anthropic/OpenAI)	API key	Per-token
CLI wrapper (this project)	OAuth subscription	Monthly flat rate

Setup Workflow

Prerequisites

• Bun 1.2 or higher
• Claude Code CLI installed and authenticated, or Codex CLI installed and authenticated
• Node.js is NOT required — Bun runs everything

Step 1: Clone and Install

git clone https://github.com/Shreyas-Dayal/ai-cli-bridge.git
cd ai-cli-bridge
bun install
cp .env.example .env
```bash

### Step 2: Configure the Admin Key

Edit the `.env` file and set `BRIDGE_ADMIN_KEY` to a strong random value. This key lets you access the admin dashboard and create per-user API keys.

```bash
BRIDGE_ADMIN_KEY=your-strong-random-key-here
PORT=3456
```bash

### Step 3: Start the Server

```bash
bun dev
```bash

The server starts on `http://localhost:3456`. For a production deployment, the README includes a Docker Compose template and a systemd service with Cloudflare Tunnel support.

### Step 4: Verify

```bash
curl http://localhost:3456/health
# → {"status":"ok"}
```bash

### Step 5: Create an API Key

Open the admin dashboard at `http://localhost:3456/admin` and generate a per-user API key. Each key can have independent limits on requests per day, requests per month, tokens per month, cost per day, and cost per month.

## Usage Examples

### Generate with Claude Code

```bash
curl -X POST http://localhost:3456/generate \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"systemPrompt":"Reply concisely.","userPrompt":"What is 2+2?"}'
```bash

### Generate with Codex CLI

```bash
curl -X POST http://localhost:3456/generate-codex \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"systemPrompt":"Reply concisely.","userPrompt":"What is 2+2?"}'

Features

• Two AI providers — Claude Code CLI (/generate) and Codex CLI (/generate-codex) behind a single API
• Per-user API keys — SHA-256 hashed, shown once at creation, timing-safe authentication
• Per-key rate limits — granular controls over requests/day, requests/month, tokens/month, cost/day, cost/month
• Admin dashboard — manage keys, monitor usage, view request logs
• Usage tracking — in-memory with periodic disk flush and auto-pruning
• Security posture — execFile (no shell injection), CSP headers, HSTS, rate limiting, input validation
• Deploy anywhere — Docker Compose support, systemd service template, Cloudflare Tunnel configuration

Practical Evaluation Checklist

• Bun is the only runtime required (no Node.js, no Docker for local dev)
• API keys are one-way hashed — the raw key is shown only once at creation
• The /health endpoint requires no auth, useful for load balancer health checks
• Rate limits are enforced per API key, not per IP — useful for multi-user shared deployments
• Both Claude Code and Codex CLI must be installed and authenticated on the host; the bridge does not store credentials

Security Notes

ai-cli-bridge uses execFile rather than exec or spawn with shell interpretation, eliminating shell injection risk. The README explicitly recommends running the server on an isolated network segment and using a Cloudflare Tunnel or VPN for any production exposure.

The admin dashboard is protected by the BRIDGE_ADMIN_KEY but does not appear to support two-factor authentication. Treat the admin key as a high-privilege secret.

FAQ

Q: Does this work with the free Claude tier? A: No. Claude Code CLI requires a paid subscription (Claude Max or equivalent). The bridge simply exposes an already-authenticated CLI as an API.

Q: Can I use both Claude and Codex on the same server? A: Yes. The server handles both providers simultaneously. Call /generate for Claude Code CLI and /generate-codex for Codex CLI.

Q: Does the bridge support streaming responses? A: The README does not explicitly confirm streaming support. Check the package.json or the source for SSE capability before relying on it.

Q: How does billing tracking work? A: Usage is tracked in-memory with periodic flushes to disk. For high-throughput production use, you may want to add an external metrics sink.

Source and Accuracy Notes

• Project page: github.com/Shreyas-Dayal/ai-cli-bridge
• License: MIT (verified via LICENSE file in repository)
• HN launch thread: news.ycombinator.com/item?id=47191530