Cliproxyapi Cloudflare Stack

CLIProxyAPI Cloudflare Stack

Skill by ara.so — Devtools Skills collection.

What It Does

This project provides a complete self-hosted stack for running an OpenAI-compatible LLM API gateway with:

• CLIProxyAPI: Multi-credential pooling with hot-reload and automatic failover
• Cloudflare Worker + D1: Programmatic email receiving for credential verification
• Cloudflare Tunnel: Zero-port-forwarding HTTPS exposure of internal gateway
• Unified Interface: Single /v1/* endpoint compatible with OpenAI SDK

Use case: Run your own LLM API proxy without sharing credentials with third-party services, supporting multiple provider accounts with automatic rotation.

Architecture Overview

[Client] 
  ↓ HTTPS
[Cloudflare Tunnel] → public domain
  ↓
[CLIProxyAPI Gateway] → localhost:5000
  ↓
[Multiple LLM Providers] (Grok, OpenAI, etc.)

[Email] → Cloudflare Email Routing → Worker → D1 Database

Installation

Prerequisites

• Cloudflare account with domain
• VM or server with Docker (for CLIProxyAPI)
• Node.js 18+ (for Wrangler CLI)
• cloudflared binary

1. Deploy CLIProxyAPI Gateway

# Clone the repository
git clone https://github.com/xsser/cliproxyapi-cloudflare-stack.git
cd cliproxyapi-cloudflare-stack

# Deploy to VM (Ubuntu/Debian)
bash scripts/deploy_vm.sh

# The script will:
# - Install Docker if needed
# - Pull CLIProxyAPI image
# - Generate API key (sk-xxxxx)
# - Create auth directory
# - Start container on port 5000

Expected output:

✓ CLIProxyAPI deployed
✓ API Key: sk-1234567890abcdef
✓ Auth directory: /opt/cpa/auth
✓ Service running on localhost:5000

Save the sk- key — this is your gateway authentication token.

2. Set Up Email Worker

# Install Wrangler
npm install -g wrangler

# Login to Cloudflare
wrangler login

# Copy and configure template
cp templates/wrangler.toml.template wrangler.toml

# Edit wrangler.toml
nano wrangler.toml

wrangler.toml configuration:

name = "email-receiver"
main = "src/index.js"
compatibility_date = "2024-01-01"

[[d1_databases]]
binding = "DB"
database_name = "email_codes"
database_id = "YOUR_D1_DATABASE_ID"

[vars]
ALLOWED_DOMAINS = "x.ai,openai.com"

Create D1 database:

# Create database
wrangler d1 create email_codes

# Note the database_id from output and update wrangler.toml

# Initialize schema
wrangler d1 execute email_codes --file=src/schema.sql

Deploy worker:

wrangler deploy

# Output will show: https://email-receiver.<your-account>.workers.dev

Configure Cloudflare Email Routing:

1. Go to Cloudflare Dashboard → Email Routing
2. Add catch-all route: *@yourdomain.com → Worker email-receiver
3. Verify domain email routing is enabled

3. Add Credentials

Credentials are JSON files in the auth directory with provider-specific format.

Example Grok credential (grok_account1.json):

{
  "provider": "grok",
  "api_key": "grok-xxxxxxxxxxxxx",
  "email": "your-email@domain.com",
  "status": "active",
  "rate_limit": {
    "requests_per_minute": 60
  }
}

Example OpenAI credential (openai_account1.json):

{
  "provider": "openai",
  "api_key": "sk-proj-xxxxxxxxxxxxx",
  "organization_id": "org-xxxxxxxxxxxxx",
  "status": "active"
}

Sync credentials to VM:

# From local directory
SRC=./auth_local VM_HOST=user@yourvm.com bash scripts/sync_credentials.sh

# Or directly on VM
scp credentials/*.json user@yourvm:/opt/cpa/auth/

CLIProxyAPI hot-reloads credentials automatically (checks every 30s).

4. Expose with Cloudflare Tunnel

# Install cloudflared
wget https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64.deb
sudo dpkg -i cloudflared-linux-amd64.deb

# Authenticate
cloudflared tunnel login

# Create tunnel
cloudflared tunnel create cpa-gateway

# Note the tunnel ID from output

# Copy service template
sudo cp templates/cloudflared-cpa.service.template /etc/systemd/system/cloudflared-cpa.service

# Edit service file
sudo nano /etc/systemd/system/cloudflared-cpa.service

cloudflared-cpa.service configuration:

[Unit]
Description=Cloudflare Tunnel for CPA Gateway
After=network.target

[Service]
Type=simple
User=cloudflared
ExecStart=/usr/local/bin/cloudflared tunnel --no-autoupdate run --token YOUR_TUNNEL_TOKEN
Restart=always
RestartSec=5

[Install]
WantedBy=multi-user.target

Create tunnel config:

mkdir -p ~/.cloudflared
nano ~/.cloudflared/config.yml

config.yml:

tunnel: YOUR_TUNNEL_ID
credentials-file: /home/user/.cloudflared/YOUR_TUNNEL_ID.json

ingress:
  - hostname: api.yourdomain.com
    service: http://localhost:5000
  - service: http_status:404

Start tunnel:

sudo systemctl daemon-reload
sudo systemctl enable cloudflared-cpa
sudo systemctl start cloudflared-cpa
sudo systemctl status cloudflared-cpa

Add DNS record in Cloudflare:

Type: CNAME
Name: api
Target: YOUR_TUNNEL_ID.cfargotunnel.com
Proxy: Yes (orange cloud)

Usage

API Endpoint

Your gateway is now available at: https://api.yourdomain.com/v1

Using with OpenAI SDK

Python:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.yourdomain.com/v1",
    api_key=os.environ.get("CPA_API_KEY")  # The sk- key from deployment
)

response = client.chat.completions.create(
    model="grok-beta",
    messages=[
        {"role": "user", "content": "Hello, Grok!"}
    ]
)

print(response.choices[0].message.content)

JavaScript/TypeScript:

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.yourdomain.com/v1',
  apiKey: process.env.CPA_API_KEY
});

const response = await client.chat.completions.create({
  model: 'grok-beta',
  messages: [
    { role: 'user', content: 'Hello, Grok!' }
  ]
});

console.log(response.choices[0].message.content);

cURL:

curl https://api.yourdomain.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $CPA_API_KEY" \
  -d '{
    "model": "grok-beta",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Key Commands

CLIProxyAPI Management

# View logs
docker logs -f cpa-gateway

# Restart gateway
docker restart cpa-gateway

# Update to latest
docker pull cliproxyapi/cliproxyapi:latest
docker restart cpa-gateway

# Check credential status
curl http://localhost:5000/admin/credentials \
  -H "Authorization: Bearer $CPA_API_KEY"

# Health check
curl http://localhost:5000/health

Credential Sync Script

# Sync from local to VM
SRC=./credentials VM_HOST=user@vm.com bash scripts/sync_credentials.sh

# Dry run (show what would be copied)
DRY_RUN=1 SRC=./credentials VM_HOST=user@vm.com bash scripts/sync_credentials.sh

# Custom auth directory
SRC=./credentials VM_HOST=user@vm.com AUTH_DIR=/custom/path bash scripts/sync_credentials.sh

Email Worker Queries

# Query D1 for received codes
wrangler d1 execute email_codes --command="SELECT * FROM verification_codes ORDER BY received_at DESC LIMIT 10"

# Clear old codes (older than 1 hour)
wrangler d1 execute email_codes --command="DELETE FROM verification_codes WHERE received_at < datetime('now', '-1 hour')"

# View worker logs
wrangler tail

Tunnel Management

# Check tunnel status
sudo systemctl status cloudflared-cpa

# View tunnel logs
sudo journalctl -u cloudflared-cpa -f

# Restart tunnel
sudo systemctl restart cloudflared-cpa

# List all tunnels
cloudflared tunnel list

# Delete tunnel
cloudflared tunnel delete cpa-gateway

Configuration

CLIProxyAPI Environment Variables

# In deploy_vm.sh or docker-compose.yml
CPA_PORT=5000                    # Gateway port
CPA_AUTH_DIR=/opt/cpa/auth      # Credentials directory
CPA_LOG_LEVEL=info              # debug, info, warn, error
CPA_POOL_SIZE=10                # Max concurrent requests per credential
CPA_RETRY_ATTEMPTS=3            # Failed request retries
CPA_HEALTH_CHECK_INTERVAL=60    # Credential health check (seconds)

Worker Environment Variables

Edit wrangler.toml:

[vars]
ALLOWED_DOMAINS = "x.ai,openai.com,anthropic.com"
MAX_CODE_AGE_MINUTES = 10
AUTO_CLEANUP_ENABLED = "true"

Rate Limiting

Configure per-credential in JSON:

{
  "provider": "grok",
  "api_key": "...",
  "rate_limit": {
    "requests_per_minute": 60,
    "tokens_per_minute": 100000,
    "requests_per_day": 5000
  }
}

Common Patterns

Multi-Provider Setup

# auth directory structure
/opt/cpa/auth/
  ├── grok_account1.json
  ├── grok_account2.json
  ├── openai_personal.json
  ├── openai_work.json
  └── anthropic_main.json

Each credential is automatically pooled. CLIProxyAPI routes based on model name:

• grok-* → Grok credentials
• gpt-* → OpenAI credentials
• claude-* → Anthropic credentials

Automatic Failover

# Client code doesn't change - failover is automatic
client = OpenAI(
    base_url="https://api.yourdomain.com/v1",
    api_key=os.environ.get("CPA_API_KEY")
)

# If grok_account1 fails, CLIProxyAPI automatically tries grok_account2
response = client.chat.completions.create(
    model="grok-beta",
    messages=[{"role": "user", "content": "Test"}]
)

Streaming Responses

stream = client.chat.completions.create(
    model="grok-beta",
    messages=[{"role": "user", "content": "Write a story"}],
    stream=True
)

for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end='')

Monitoring Credential Usage

# Check logs for credential selection
docker logs cpa-gateway 2>&1 | grep "Using credential"

# Example output:
# [INFO] Using credential grok_account1 for model grok-beta
# [INFO] Request successful with grok_account1 (245ms)
# [WARN] Credential grok_account1 rate limited, trying grok_account2

Troubleshooting

Gateway Returns 502 Bad Gateway

Cause: CLIProxyAPI container not running or no active credentials.

Fix:

# Check container
docker ps | grep cpa-gateway

# View logs
docker logs cpa-gateway

# Verify credentials exist
ls -la /opt/cpa/auth/

# Restart
docker restart cpa-gateway

"No active credentials for provider" Error

Cause: No valid credential JSON files for requested model's provider.

Fix:

# Check credential format
cat /opt/cpa/auth/grok_account1.json

# Must have:
# - "provider" field matching model prefix
# - "api_key" field
# - "status": "active"

# Force credential reload
docker restart cpa-gateway

Email Worker Not Receiving Codes

Cause: Email routing not configured or D1 binding missing.

Fix:

# Test email routing
echo "test" | mail -s "Test" test@yourdomain.com

# Check worker logs
wrangler tail

# Verify D1 binding
wrangler d1 info email_codes

# Redeploy worker
wrangler deploy

Tunnel Connection Issues

Cause: Tunnel service not running or incorrect credentials.

Fix:

# Check tunnel status
sudo systemctl status cloudflared-cpa

# Test local gateway
curl http://localhost:5000/health

# Verify tunnel config
cat ~/.cloudflared/config.yml

# Check DNS
nslookup api.yourdomain.com

# Restart tunnel
sudo systemctl restart cloudflared-cpa

Rate Limit Exceeded

Cause: All credentials in pool exhausted rate limits.

Fix:

# Add more credentials to pool
scp new_credential.json user@vm:/opt/cpa/auth/

# Adjust rate limits in credential JSON
nano /opt/cpa/auth/grok_account1.json
# Increase "requests_per_minute"

# Wait for rate limit reset (typically 1 minute)

Invalid API Key Response

Cause: Using wrong API key or credential expired.

Fix:

# Verify you're using the gateway key (sk-), not provider key
echo $CPA_API_KEY

# Check credential validity
docker logs cpa-gateway | grep "auth failed"

# Update expired credential
nano /opt/cpa/auth/expired_credential.json
# Update api_key or set "status": "inactive"

High Latency

Cause: VM resources constrained or network issues.

Fix:

# Check CPU/memory
htop

# Check network latency to providers
ping api.x.ai
ping api.openai.com

# Increase pool size for better concurrency
# Edit docker run command in deploy_vm.sh:
-e CPA_POOL_SIZE=20

# Redeploy
bash scripts/deploy_vm.sh

Security Best Practices

1. Never commit credentials: Keep .json files in .gitignore
2. Rotate gateway key: Regenerate sk- key periodically
3. Use environment variables: Store all secrets in env vars, not code
4. Enable Cloudflare WAF: Add rate limiting and bot protection to tunnel domain
5. Monitor usage: Set up alerts for abnormal API usage patterns
6. Restrict tunnel access: Use Cloudflare Access for additional auth layer

Advanced: Custom Credential Providers

Create custom credential format for new providers:

{
  "provider": "custom-llm",
  "api_key": "...",
  "endpoint": "https://api.custom-llm.com/v1",
  "model_prefix": "custom-",
  "status": "active",
  "rate_limit": {
    "requests_per_minute": 100
  }
}

CLIProxyAPI will route any model starting with custom- to this credential's endpoint.