When I built FreeAPI (https://findfreeapi.com/en), a catalog of 1,047 public APIs with an in-browser test runner, the hardest part was the proxy. Letting anyone send requests through your server to arbitrary URLs is a classic SSRF vector.
Simple URL blocklists fail against DNS rebinding (TOCTOU) and alternate IP encodings like bracketed IPv6 or decimal IPs. To handle that, I wrote a pinned-IP fetch wrapper that resolves DNS upfront, rejects private or reserved ranges, and connects directly to the validated IP.
Three other guardrails kept the server stable under anonymous traffic:
- A strict 5 MB payload ceiling so a huge upstream file cannot exhaust memory.
- An SHA-256 query-param cache and circuit breaker to avoid hammering slow third-party hosts.
- A 60 requests per minute rate limit per IP for anonymous sandbox users.
Another surprise was how often public APIs return HTTP 200 with an HTML bot-block page instead of JSON. Instead of passing that back as a 502 server error, the proxy inspects the content type and returns a 422 so the UI can explain what happened.
The catalog itself is statically generated across 1,300+ Next.js 15 pages, with 609 no-auth APIs and 33 client-side utilities. If you have built a similar user-facing proxy, I am curious what edge cases bit you in production.
Top comments (0)