API

Every JSON answer is { success, data, timestamp }. Handlers live in server/api/ and server/routes/, and this table is the list in shared/endpoints.ts.

Endpoints

Endpoint	Returns
`GET /api/daily-word` `?date=YYYY-MM-DD`	One entry plus its `prev` and `next` days and `sameDay`, the open words of the same month and day in other years, oldest first. With no date: today in Japan; once the catalogue runs out the words start again from the first, and `lap` says which pass it is (1 until then, always 1 for an explicit date). A future or invalid date is a `400`; a past date with no entry (or no word yet) is a `404`. `next` stays `null` until that day has arrived.
`GET /api/word-calendar` `?month=YYYY-MM&q=&level=&stratum=&process=&pos=&frequency=&pitch=&part=&match=`	`{ month, months, today, days, filters, total, count, monthCounts, facets }`. An open day carries its `term`, `kana`, `stratum` and `match` (whether it passes the filters, always true with none); an upcoming day carries only its date and `"upcoming"`, and is never matched. Once the catalogue has run out, an open day after the last entry shows the word of its lap and also carries `wordDate` (the day that word first opened) and `lap`, and `months` runs on to the current month. The filters are the `/api/explore` ones, with the same counts: `count` of `total` open words match across every month, `monthCounts` gives each month's matches and `facets` the options' counts. The month defaults to the current one if it has words, else the newest. A malformed month or an invalid filter is a `400`, a month with no words a `404`.
`GET /api/parts`	The parts index, `{ parts: [{ text, count, readings }] }`: every morpheme an open word's “Taken apart” row shows, most-used first.
`GET /api/part` `?text=日`	One part with the open words that show it, grouped by the reading it has in each: `{ text, count, readings, alsoIn }`. Only days that have arrived count, so a part seen only in an upcoming word is a `404`. A missing or over-long `text` is a `400`.
`GET /api/kanji`	The kanji index, `{ kanji: [{ char, count, grade? }] }`: every kanji an open word is written with, most-used first, with KANJIDIC2's school grade.
`GET /api/kanji-detail` `?char=日`	KANJIDIC2's record of one kanji with the open words written with it: `{ char, strokeCount, grade?, freq?, on, kun, meanings, count, words, isPart, strokes?, components? }`, where each of `words` carries its JLPT `level` and `strokes` is KanjiVG's stroke paths in drawing order on a 109 by 109 square, left out when KanjiVG and KANJIDIC2 count the strokes differently, and `components` is the shapes KRADFILE lists it as containing, each with a few other open kanji that contain it, left out where it lists none. Only days that have arrived count, so a kanji used only by an upcoming word is a `404`. A missing value or more than one character is a `400`.
`GET /api/explore` `?q=&level=&stratum=&process=&pos=&frequency=&pitch=&part=&match=&limit=`	The open words matching every filter, newest first: `{ filters, total, count, words, facets }`. `count` is every match, but `words` holds only the newest 100 unless `limit` asks for more (a whole number of at least 1). Every filter is optional and an empty value means “no filter”; anything else invalid is a `400`. `level`, `stratum` (which also takes `unstated`), `process`, `pos` (a group such as `verb`), `frequency` (`common`, `less` or `unlisted`, from JMdict's priority codes) and `pitch` (`heiban`, `atamadaka`, `nakadaka`, `odaka` or `unstated`, the patterns of the accents the entry shows) take several comma-joined choices; `match=all` makes a word carry every process, part of speech and accent pattern chosen instead of any. Each facet counts the words the other filters leave.
`GET /api/patterns`	Counts across the open words: `{ total, withParts, withBase, strata, levels, processes, pairs, combinations, rendaku, pitch, lengths, soundAlikes }`, the last three being the pitch accents split into patterns by layer, the mean length of the words by level, and the readings two or more words share (`{ readings, words, withAccents, toldApart, groups }`, the readings with the most words in `groups`).
`GET /api/related` `?date=YYYY-MM-DD`	Up to six open words that resemble one entry (shared parts, processes or layer), closest first, each with what it shares; the words JMdict points the entry's sense at (`antonyms`, `seeAlso`), each with the open word it is where there is exactly one; and the open words with the same reading and another spelling, oldest first, each with its pitch accents (`soundAlikes`): `{ date, words, antonyms, seeAlso, soundAlikes }`. `date` is required; a future or malformed one is a `400`, a day with no entry a `404`.
`GET /api/catalogue`	`{ first, last, total, open, coverage }`: the first and last day of the written catalogue, how many words it holds and how many have opened, and in `coverage` how many of the open words carry a layer, a pitch accent, example sentences, parts, processes and a quoted origin. The word range and the coverage table quoted in the docs come from this.
`GET /sitemap.xml`	Server route, not under `/api`. Lists the static pages, every open word and each part or kanji seen in more than one open word, never an upcoming day. URLs use `NUXT_PUBLIC_SITE_URL` when set, else the request's own origin.
`GET /og.png` `?date=YYYY-MM-DD`	Server route, not under `/api` (so `robots.txt` does not block link previews). The share image of one open word as a 1200 by 630 PNG: the word, its reading, meaning, level, layer and headline, in the palette of its day's season. A future, malformed or unknown date is a `404`, so the image cannot reveal a word early. The page's Open Graph tags point here.
`GET /robots.txt`	Server route, not under `/api`. Disallows `/api/` and names the sitemap.

A successful answer is cached until midnight in Japan (Caching). An address that asks an /api/ endpoint or the share image more than 120 times in 60 seconds gets a 429 with Retry-After. The count is kept in memory, never stored, so each server instance counts alone, and a client behind a proxy that does not set X-Forwarded-For is counted by the proxy's address. A day's words are served only once it is open (Daily words); what each failure looks like on a page is in Error and fallback states.

Example

GET /api/daily-word?date=2013-05-04, abridged

{
  "success": true,
  "data": {
    "entry": {
      "date": "2013-05-04",
      "term": "サンキュー",
      "kana": "サンキュー",
      "meaning": "thank you",
      "level": "N1",
      "pos": ["interjection (kandoushi)"],
      "priority": ["gai1"],
      "register": ["colloquial"],
      "pitch": [1],
      "stratum": "gairaigo",
      "processes": ["borrowing"],
      "headline": "サンキュー is borrowed from English thank you.",
      "morphemes": [
        { "text": "サンキュー", "reading": "サンキュー", "meaning": "English thank you" }
      ],
      "sources": [{ "quote": "Borrowed from English thank you." }],
      "wiktionaryDump": "2026-09-02"
    },
    "lap": 1,
    "prev": { "date": "2013-05-03", "term": "免れる" },
    "next": { "date": "2013-05-05", "term": "霞む" }
  },
  "timestamp": "2026-10-07T00:00:00.000Z"
}