Web app · case study

Next-movies

Movie and TV browser running as a Cloudflare Worker, with rate limiting and IP bans in front of the TMDB API.

My role: Sole developer
Type: Web application
Stack: Next.js, TypeScript, Tailwind CSS, Redis
Runs on: Cloudflare Workers, no origin server
Edge: Cloudflare, custom domain on the Worker

Why I built it

I built Next-movies because I watch a lot of movies and wanted my own website for looking up films, shows and actors the way I like, instead of searching the web every time.It was my first Next.js project, built while I was learning the framework.It has since become where I practise deployment: it moved from Vercel to a VPS, and now runs as a Cloudflare Worker with no origin server.

The problem

My own website for finding movies, TV shows and actors, with the filters and layout I want.
My first Next.js app, used to learn the framework.
Keep the TMDB token secret and the free API quota safe from scrapers, at no hosting cost.

Features

Infinite scrolling
In-app trailer playback

Architecture

Dev · applicationSec · security controlOps · infrastructureExternal · third partyScroll sideways to see all of it.

In words

The app runs as a Cloudflare Worker via OpenNext, with no origin server.Middleware blocks script user agents, checks the IP ban list, and rate-limits through Upstash Redis.Only then does the TMDB proxy run.The TMDB token is a Worker secret and never reaches the browser.Fetch results persist in a KV cache behind the regional Cache API.Static assets come from the Worker's assets binding; images load straight from TMDB's CDN.

Pipeline

push → main

Workers Builds
Cloudflare's own CI
opennextjs-cloudflare build
Next.js → Worker bundle
Deploy
new Worker version, atomic

Decisions

Cloudflare Workers, after Vercel and a VPS.: It moved from Vercel to a hardened VPS (Docker, nginx), then to Workers via OpenNext.The VPS setup was deleted after.
The Cache API in front of the KV cache.: Renders reached 50% of the daily KV operations cap.Cache API reads are unmetered, so repeat reads skip KV.
Upstash Redis for rate limits and bans.: A Worker runs as many instances.Only a shared store gives one global limit.A strict in-memory limit is the fallback.
Trust only the CF-Connecting-IP header.: Cloudflare sets it; clients can't forge it.X-Forwarded-For is never trusted.
Images load straight from TMDB's CDN.: No image-optimization worker to run.No image proxy route to expose.

Challenges

Free-tier limits at the edge.: The Cache API layer cut KV usage without a paid plan.
Moving a Next.js app onto the Workers runtime.: OpenNext handles the build.The health check was rewritten for the Workers runtime.

Testing

A QA script checks function, security, performance and edge cases on a running instance.
A validator tests the rate limiter in memory or against Upstash.
Next.js and sharp advisories are patched as they appear.
Not yet: either script in CI.

Outcome

Live at nextmovies.wyco-dev.com, served from Cloudflare's edge.

Run the QA script after every deploy
Accounts and watchlists

Stack

Next.js: Runs as a Cloudflare Worker via OpenNext.Renders pages and proxies the movie API.
TypeScript: Types the movie, show, and actor payloads.
Tailwind CSS: Styles the browse and detail views.
Redis: Rate limits, IP bans and list caching, on Upstash.

Security

Authentication: None.Public read-only app.
Input validation: Allow-listed API keys.Every query parameter is validated.
Protections: The API proxy keeps the token server-side.Rate limits, IP bans and security headers apply.

Operations

Runs on: Cloudflare Workers, no origin server
Edge: Cloudflare, custom domain on the Worker
Deploy: Push to main triggers Workers Builds.It runs the OpenNext build and ships a new Worker version.The TMDB token is a Worker secret.
Health check: Middleware returns 403 to scripts and 200 to browsers.A 403 to a script means the filter works.