7.1 KiB
MyAnimeList
|
MyAnimeList My personal anime tracker, built because nothing else felt right. |
Why this project exists
I built this for myself.
I was frustrated with the UI and UX of every tracker I tried. Even when something looked decent, it still felt awkward to use day-to-day, or it was missing pieces I considered essential. I wanted one place that matched how I actually watch anime: search fast, get context fast, update status fast, and move on.
So this project is personal first and public second. I put it on GitHub because I like shipping in the open, not because it was originally designed as a general-purpose product for everyone.
Technically, I also wanted to prove that a small, server-rendered Go app could stay reliable even when upstream anime APIs are inconsistent. A lot of this code exists because real APIs rate-limit, timeout, and occasionally fail at the worst possible moment.
What the application offers
For my own workflow, MyAnimeList combines catalog browsing, seasonal discovery, quick search, detail pages with recommendations and relations, and full watchlist management in one server-rendered interface. It also includes account flows such as registration, login, recovery, and recovery-key rotation. The notifications area is tuned for practical tracking, including sequel visibility derived from watchlist context.
Technical approach
The application is written in Go and rendered on the server with templ, with SQLite as the primary datastore and sqlc for typed query generation. HTMX and small JavaScript modules handle incremental interactions, which keeps the interface responsive without moving the entire product into a heavy client-side architecture.
The external anime data source is Jikan (https://api.jikan.moe/v4). Because reliability is a first-class concern, the client layer includes request pacing, bounded retries, backoff behavior, stale-cache fallback, and a persisted retry queue for failed fetches that should be retried later.
Repository structure
Instead of treating the repository as one flat service, the codebase is organized into focused boundaries.
| Path | Purpose |
|---|---|
cmd/server |
Application entrypoint and process lifecycle setup |
internal/server |
Route registration and middleware composition |
internal/features/anime |
Anime browsing, discovery, search, detail, and notifications logic |
internal/features/watchlist |
Watchlist updates, retrieval, import, and export |
internal/features/auth |
Authentication, sessions, account recovery, and account settings |
internal/jikan |
Upstream API client, caching, and retry-aware fetch behavior |
internal/worker |
Background relation sync, retry processing, and cache cleanup |
internal/database |
Migration runner, generated query layer, and DB models |
internal/templates |
Server-rendered page and partial templates |
migrations |
Schema evolution and operational DB changes |
static |
CSS, JavaScript, and static assets |
Runtime behavior
On startup, the server opens SQLite using DATABASE_FILE (defaulting to mal.db), runs migrations automatically, initializes core services, starts the background worker, and then serves HTTP traffic on PORT (defaulting to 3000). A request enters the router, passes through global middleware for origin and authentication boundaries, reaches a feature handler, and then resolves through service logic that combines database access with upstream data where needed before rendering HTML.
The background worker continuously maintains relation data for sequel awareness, processes queued retryable anime fetches, and periodically removes expired cache records. This keeps user-facing pages stable even when data collection has to happen in multiple phases.
Reliability and tradeoffs
The hardest part has been balancing freshness and resilience. Upstream APIs can fail transiently with 429 and 5xx responses, so the app favors graceful degradation over hard failure. Cached values are used when fresh requests fail, retryable failures are persisted and replayed in the worker, and relation synchronization is incremental so one bad fetch does not block the rest of the graph.
There are still honest limits. Metadata quality still depends partly on external providers, and there is also no formal CI pipeline yet, so local validation is currently the main quality gate.
Getting started
For local development, install Go 1.24+, SQLite, Bun, and the templ CLI, then generate templates, build frontend assets, and run the server.
go install github.com/a-h/templ/cmd/templ@latest
bun install
templ generate
bun run build:assets
go run ./cmd/server
The frontend pipeline uses a single source stylesheet (static/style.css) and TypeScript sources in static/*.ts, then emits build artifacts (static/tailwind.css and static/*.js) for serving.
When the server starts, the app is available at http://localhost:3000.
For containerized usage, the included Dockerfile uses a multi-stage build that generates templates, compiles cmd/server, and ships a slim runtime image with SQLite support.
docker build -t myanimelist .
docker run --rm -p 3000:3000 myanimelist
Configuration
| Variable | Default | Description |
|---|---|---|
PORT |
3000 |
HTTP listen port |
DATABASE_FILE |
mal.db |
SQLite database file path |
ENV |
(empty) | Set to production to enable secure session cookies |
Database and testing
Migrations run at startup, so schema changes are applied automatically before the server begins accepting traffic. Migration history includes the initial auth and watchlist schema, anime metadata expansion, relation tracking, Jikan cache persistence, indexing updates, recovery key support, and retry-queue support for failed fetches.
Tests are available for watchlist behavior, relation helpers, auth middleware boundaries, and watch-order parsing. Run the full test suite with:
go test ./...
Contributing
This is primarily a personal project, so development priorities are driven by my own use and preferences. That said, if you spot a bug or have a focused improvement, feel free to open an issue or pull request. Please read CONTRIBUTING.md first so expectations around scope, validation, and security handling are clear.
Security
Keep secrets out of version control, do not publish real credentials in documentation or screenshots, and report security issues privately before public disclosure.
License
This project is released under the MIT License. See LICENSE for details.