tavily-key-generator-proxy
Tavily Key Generator + API Proxy
Skill by ara.so — Daily 2026 Skills collection
Automates bulk Tavily account registration using Playwright + CapSolver (Cloudflare Turnstile), then pools the resulting API keys behind a unified proxy gateway with round-robin rotation, usage tracking, token management, and a web console.
What It Does
| Component | Location | Purpose |
|---|---|---|
| Key Generator | root / |
Playwright-driven headless Firefox registers Tavily accounts, solves Turnstile CAPTCHAs, verifies email, extracts API key |
| API Proxy | proxy/ |
FastAPI service that round-robins requests across pooled keys, exposes /api/search and /api/extract, serves web console at /console |
Each free Tavily account yields 1,000 API calls/month. The proxy aggregates quota: 10 keys = 10,000 calls/month via one endpoint.
Installation
Key Generator
git clone https://github.com/skernelx/tavily-key-generator.git
cd tavily-key-generator
pip install -r requirements.txt
playwright install firefox
cp config.example.py config.py
# Edit config.py with your CapSolver key and email backend
python main.py
API Proxy (Docker — recommended)
cd proxy/
cp .env.example .env
# Set ADMIN_PASSWORD in .env
docker compose up -d
# Console at http://localhost:9874/console
Configuration (config.py)
CAPTCHA Solver (required)
# CapSolver — recommended (~$0.001/solve, high success rate)
CAPTCHA_SOLVER = "capsolver"
CAPSOLVER_API_KEY = "CAP-..." # from capsolver.com
# Browser click fallback — free but low success rate
CAPTCHA_SOLVER = "browser"
Email Backend (required — pick one)
Option A: Cloudflare Email Worker (self-hosted, free)
EMAIL_BACKEND = "cloudflare"
EMAIL_DOMAIN = "mail.yourdomain.com"
EMAIL_API_URL = "https://mail.yourdomain.com"
EMAIL_API_TOKEN = "your-worker-token"
Option B: DuckMail (third-party temporary email)
EMAIL_BACKEND = "duckmail"
DUCKMAIL_API_BASE = "https://api.duckmail.sbs"
DUCKMAIL_BEARER = "dk_..."
DUCKMAIL_DOMAIN = "duckmail.sbs"
If both backends are configured, the CLI prompts you to choose at runtime.
Registration Throttle (anti-ban)
THREADS = 2 # Max 3 — more = higher ban risk
COOLDOWN_BASE = 45 # Seconds between registrations
COOLDOWN_JITTER = 15 # Random additional seconds
BATCH_LIMIT = 20 # Pause after this many registrations
Auto-upload to Proxy (optional)
PROXY_AUTO_UPLOAD = True
PROXY_URL = "http://localhost:9874"
PROXY_ADMIN_PASSWORD = "your-admin-password"
Running the Key Generator
python main.py
# Prompts: how many keys to generate, which email backend
# Output: api_keys.md with all generated keys
# If PROXY_AUTO_UPLOAD=True, keys are pushed to proxy automatically
Generated api_keys.md format (used for bulk import into proxy):
tvly-abc123...
tvly-def456...
tvly-ghi789...
API Proxy Usage
Calling the Proxy (drop-in Tavily replacement)
# Search — replace base URL only
curl -X POST http://your-server:9874/api/search \
-H "Authorization: Bearer tvly-YOUR_PROXY_TOKEN" \
-H "Content-Type: application/json" \
-d '{"query": "latest AI news", "search_depth": "basic"}'
# Extract
curl -X POST http://your-server:9874/api/extract \
-H "Authorization: Bearer tvly-YOUR_PROXY_TOKEN" \
-H "Content-Type: application/json" \
-d '{"urls": ["https://example.com/article"]}'
Token can also be passed in the request body as api_key (Tavily SDK compatible):
import requests
response = requests.post(
"http://your-server:9874/api/search",
json={
"query": "Python web scraping",
"api_key": "tvly-YOUR_PROXY_TOKEN"
}
)
print(response.json())
Python SDK Integration
from tavily import TavilyClient
# Point SDK at your proxy — no other changes needed
client = TavilyClient(
api_key="tvly-YOUR_PROXY_TOKEN",
base_url="http://your-server:9874"
)
results = client.search("machine learning trends 2025")
Admin API Reference
All admin endpoints require header: X-Admin-Password: your-password
Keys Management
# List all keys
curl http://localhost:9874/api/keys \
-H "X-Admin-Password: secret"
# Add a single key
curl -X POST http://localhost:9874/api/keys \
-H "X-Admin-Password: secret" \
-H "Content-Type: application/json" \
-d '{"key": "tvly-abc123..."}'
# Bulk import from api_keys.md text
curl -X POST http://localhost:9874/api/keys \
-H "X-Admin-Password: secret" \
-H "Content-Type: application/json" \
-d '{"bulk": "tvly-abc...\ntvly-def...\ntvly-ghi..."}'
# Toggle key enabled/disabled
curl -X PUT http://localhost:9874/api/keys/{id}/toggle \
-H "X-Admin-Password: secret"
# Delete key
curl -X DELETE http://localhost:9874/api/keys/{id} \
-H "X-Admin-Password: secret"
Token Management
# Create access token
curl -X POST http://localhost:9874/api/tokens \
-H "X-Admin-Password: secret" \
-H "Content-Type: application/json" \
-d '{"label": "my-app"}'
# Returns: {"token": "tvly-...", "id": "..."}
# List tokens
curl http://localhost:9874/api/tokens \
-H "X-Admin-Password: secret"
# Delete token
curl -X DELETE http://localhost:9874/api/tokens/{id} \
-H "X-Admin-Password: secret"
Stats
curl http://localhost:9874/api/stats \
-H "X-Admin-Password: secret"
# Returns: total_quota, used, remaining, active_keys, disabled_keys
Change Admin Password
curl -X PUT http://localhost:9874/api/password \
-H "X-Admin-Password: current-password" \
-H "Content-Type: application/json" \
-d '{"new_password": "new-secure-password"}'
Proxy Behavior
- Round-robin: requests distributed evenly across active keys
- Auto-disable: key disabled after 3 consecutive failures
- Quota tracking: total = active_keys × 1,000/month; updates automatically on add/delete/toggle
- Token format: proxy tokens use
tvly-prefix, indistinguishable from real Tavily keys to clients
Proxy .env Configuration
ADMIN_PASSWORD=change-this-immediately
PORT=9874
# Optional: restrict CORS origins
CORS_ORIGINS=https://myapp.com,https://app2.com
Nginx Reverse Proxy + HTTPS
server {
listen 443 ssl;
server_name tavily-proxy.yourdomain.com;
ssl_certificate /etc/letsencrypt/live/yourdomain.com/fullchain.pem;
ssl_certificate_key /etc/letsencrypt/live/yourdomain.com/privkey.pem;
location / {
proxy_pass http://127.0.0.1:9874;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
}
}
Integration Pattern: Auto-replenish Pool
# replenish.py — run on cron when quota runs low
import requests
PROXY_URL = "http://localhost:9874"
ADMIN_PASSWORD = "your-password"
MIN_ACTIVE_KEYS = 10
def get_stats():
r = requests.get(f"{PROXY_URL}/api/stats",
headers={"X-Admin-Password": ADMIN_PASSWORD})
return r.json()
def trigger_registration(count: int):
"""Call your key generator as subprocess or import directly"""
import subprocess
subprocess.run(["python", "main.py", "--count", str(count)], check=True)
stats = get_stats()
active = stats["active_keys"]
if active < MIN_ACTIVE_KEYS:
needed = MIN_ACTIVE_KEYS - active
print(f"Pool low ({active} keys), registering {needed} more...")
trigger_registration(needed)
Troubleshooting
| Problem | Cause | Fix |
|---|---|---|
| CAPTCHA solve failures | Tavily elevated bot detection | Pause several hours, reduce THREADS to 1 |
| Email verification timeout | Slow email delivery | Increase poll timeout in config |
| Keys immediately disabled in proxy | Tavily account flagged/suspended | Register from different IP |
playwright install firefox fails |
Missing system deps | Run playwright install-deps firefox first |
| Docker compose port conflict | 9874 in use | Change PORT in .env and docker-compose.yml |
X-Admin-Password 401 |
Wrong password | Check .env, restart container after changes |
| IP ban during registration | Too many registrations | Use BATCH_LIMIT=10, wait hours between batches |
Rate Limit Safe Settings
# Conservative — minimizes ban risk
THREADS = 1
COOLDOWN_BASE = 60
COOLDOWN_JITTER = 30
BATCH_LIMIT = 10
Security Checklist
- Change
ADMIN_PASSWORDfrom default immediately after deploy - Add
config.pyto.gitignore(already included, verify before push) - Deploy behind HTTPS in production
- Rotate proxy tokens periodically
- Never commit
api_keys.mdto public repositories