Status: [UNTESTED] Endpoints found: https://mydomain.ru/api/wordstat/top` Source: https://github.com/t8t7/wordstat-gpt-proxy # Wordstat Proxy API Authenticated browser bridge for the free Yandex Wordstat website. It returns the visible "Popular" and "Similar" tables as JSON and does not use the paid Yandex Search API. The service is designed for a GPT Action: deploy it behind HTTPS, import `https://YOUR_PUBLIC_HOST/openapi.json` into the GPT configuration, select API key authentication with `Bearer` auth, and enter the value of `PROXY_API_KEY`. ## Server setup 1. Copy `.env.example` to `.env` on the VPS and replace `PROXY_API_KEY`. Never commit `.env`. 2. Start the isolated stack: ```bash docker compose up -d --build docker compose ps ``` 3. Merge one block from `deploy/Caddyfile.example` into the current Caddyfile, validate it, and only then reload Caddy. Do not replace the existing file. 4. Create an SSH tunnel to the browser screen: ```bash ssh -L 7901:127.0.0.1:7901 goodpapa@YOUR_VPS_IP ``` 5. Open `http://127.0.0.1:7901/?autoconnect=1&resize=scale`, then sign in to Yandex manually. The browser profile is stored in a private Docker volume. The API and browser-screen ports are bound to loopback only. Redis has no published port. The browser screen must never be exposed directly through Caddy; access it only through SSH. ## Verification On the VPS: ```bash curl --fail --silent http://127.0.0.1:8087/health ``` Through the subdomain: ```bash curl --get 'https://wordstat.mydomain.ru/top' \ --data-urlencode 'q=купить базу клиентов' \ --header 'Authorization: Bearer YOUR_PROXY_API_KEY' ``` For the path-based Caddy variant, set `PUBLIC_BASE_URL` and use `https://mydomain.ru/api/wordstat/top` instead. Successful responses use numeric counts parsed from the visible Wordstat page: ```json { "query": "купить базу клиентов", "totalCount": 12345, "results": [ {"phrase": "купить базу клиентов", "count": 12345} ], "associations": [ {"phrase": "база потенциальных