ai-cli-bridge

Turn your Claude Max / OpenAI Pro subscriptions into a private HTTP API that any project can consume.

This server wraps the Claude Code CLI and Codex CLI behind an Express API. Instead of paying per-token, requests are backed by your existing subscription.

Why

AI API calls are expensive at scale. If you already pay for Claude Max ($100-200/mo) or OpenAI Pro ($200/mo), that usage is locked to the CLI tools — they can't be called from web apps, plugins, or other HTTP clients. This project bridges that gap.

Features

• Two providers — Claude Code CLI (/generate) and Codex CLI (/generate-codex)
• Per-user API keys — SHA-256 hashed, shown once at creation, timing-safe auth
• Per-key usage limits — 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, auto-pruning
• Security — execFile (no shell injection), CSP, HSTS, rate limiting, input validation
• Deploy anywhere — Docker support, systemd service, Cloudflare Tunnel template

Quick Start

Prerequisites

• Bun 1.2+
• Claude Code CLI and/or Codex CLI installed and authenticated

Local Development

git clone https://github.com/Shreyas-Dayal/ai-cli-bridge.git
cd ai-cli-bridge
bun install
cp .env.example .env    # Edit to set BRIDGE_ADMIN_KEY
bun dev                 # Starts on http://localhost:3456

Docker

cp .env.example .env    # Edit to set BRIDGE_ADMIN_KEY
docker compose up --build -d

Verify

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

Usage

Generate (Claude)

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

Generate (Codex)

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

Response Shape

Both endpoints return a consistent response shape:

{
  "content": [{ "type": "text", "text": "4" }],
  "usage": {
    "input_tokens": 1234,
    "output_tokens": 56
  },
  "cost_usd": 0.003
}

Claude responses also include duration_ms, cache_creation_input_tokens, and cache_read_input_tokens. Codex cost is estimated from a built-in pricing table.

Streaming

Add "stream": true to the request body to receive Server-Sent Events instead of a JSON response:

curl -N -X POST http://localhost:3456/generate \
  -H "Authorization: Bearer <USER_KEY>" \
  -H "Content-Type: application/json" \
  -d '{"systemPrompt":"Reply concisely.","userPrompt":"Write a haiku about coding.","stream":true}'

SSE events:

Streaming works on both /generate and /generate-codex. Omitting stream or setting it to false returns the standard JSON response.

API Endpoints

User (Bearer user-key)

Admin (Bearer admin-key)

Key Management

Create your first user key:

curl -X POST http://localhost:3456/admin/keys \
  -H "Authorization: Bearer <ADMIN_KEY>" \
  -H "Content-Type: application/json" \
  -d '{"name":"my-app","maxCostPerMonth":10.00}'

The raw key is returned once — save it. Only the SHA-256 hash is stored on disk.

Key names must be 1-50 characters: alphanumeric, hyphens, and underscores only. All limits default to 0 (unlimited). Available limits: maxRequestsPerDay, maxRequestsPerMonth, maxTokensPerMonth, maxCostPerDay, maxCostPerMonth.

Deployment

The recommended production setup is a cheap VPS ($4-6/mo) behind a Cloudflare Tunnel.

See GUIDE.md for a complete step-by-step deployment walkthrough, and DEPLOY.md for a quick deployment reference.

Configuration

All settings are via environment variables. See .env.example for the full list. Invalid values (e.g. PORT=abc) will cause the server to fail fast on startup with a clear error message.

Codex cost estimation uses pricing from codex-pricing.json. Edit that file when prices change — no rebuild needed, just restart. Claude CLI returns cost directly.

Security

• No shell injection — CLI invocations use execFile (array args, no shell)
• Timing-safe auth — crypto.timingSafeEqual for key comparison
• Security headers — CSP, HSTS, X-Frame-Options DENY, Referrer-Policy, Permissions-Policy
• Rate limiting — global limiter + separate admin endpoint limiter
• Input validation — key name format, non-negative limits, prompt length cap (500K chars)
• File permissions — data files written with 0o600 (owner-only)

See SECURITY.md for vulnerability reporting and deployment hardening guidance.

Project Structure

src/
  server.ts               Express app, routes, middleware
  config.ts               Environment variable parsing
  keys.ts                 Key CRUD, usage tracking, limits
  middleware/auth.ts       Per-user + admin auth (timing-safe)
  providers/
    claude.ts             Claude Code CLI wrapper
    codex.ts              Codex CLI wrapper
public/                   Admin dashboard (HTML/CSS/JS)
data/                     Runtime data (gitignored)
codex-pricing.json        Codex model pricing (editable, no rebuild)
Dockerfile                Docker image definition
docker-compose.yml        Docker orchestration
ai-cli-bridge.service     systemd service file
cloudflared-config.yml    Cloudflare Tunnel template
.github/workflows/ci.yml  CI build check on push/PR

Disclaimer

This project may violate the Terms of Service of the underlying CLI providers. You are responsible for reviewing and complying with the applicable terms before using this software.

• Anthropic — As of February 2026, Anthropic's Consumer Terms of Service prohibit using OAuth tokens from Free, Pro, or Max subscriptions in any third-party product, tool, or service. Wrapping the Claude Code CLI behind an HTTP API and sharing access with other users likely falls under this prohibition. See Anthropic's policy clarification for details.
• OpenAI — OpenAI's Terms of Use prohibit sharing account credentials and reselling access. Exposing a Codex CLI subscription as a multi-user API may constitute account sharing.

This project is provided as-is for educational and experimental purposes. The authors are not responsible for any consequences arising from its use. If the providers update their terms or enforcement, your access may be revoked without notice.

Contributing

See CONTRIBUTING.md for development setup, code style, and PR guidelines.

License

MIT