immit dictionary MCP Server

The inflection table of one English or Japanese word, grouped by form in the order a textbook table shows them: for Japanese the polite, て, past, negative and other forms, for English the third person, past, participle or plural. Each form carries its name in the word's language with English and Japanese translations, and one row per surface with the text, the polarity (positive or negative) and tone (plain or polite) on Japanese rows, the homograph it belongs to, the part of speech and the kana reading. The pitch accent and the MP3 are the headword's, on the pronunciations lookup_word answers, not each row's. Call it to say how a word is conjugated or inflected, the polite past of 帰る for example, rather than inflecting it yourself. headword resolves as lookup_word's text does. A word with no inflections, a Japanese noun or a particle, answers an empty forms list, not an error.

The immit dictionary entry for one English or Japanese word: its senses with definitions, translations and example sentence pairs, every inflected form as a row with its reading, the pronunciations with IPA for English or the kana reading and pitch accent for Japanese and a playable MP3 URL for each, the frequency rank, synonyms and antonyms, and alternative spellings. Call it rather than answering from memory whenever the question is about pitch accent, a reading, an inflection table, how a word sounds, an example sentence, how common a word is, or an exact definition: those are the facts a model guesses wrong. text is the dictionary form (食べる, not 食べた; eat, not eating); a case variant or an alternative spelling resolves to its entry, an inflected form does not, so take the dictionary form or use search_words, which matches inflected forms. The result is the entry JSON that the immit://entry-shape resource describes field by field.

Tell immit that a real word has no entry, so a lexicographer adds one. Call it when lookup_word and search_words both miss a word a person asked about and you are confident the word exists in that language: a headword in its dictionary form, never an inflected form, a typo or a proper noun. context is optional, the sentence the word came from or what the person wanted to know. The report is read by a person; nothing is added on the spot, so answer the person from what you know and say the dictionary lacks the word.

Send the people who make immit a note: an entry that looks wrong, a tool that answered badly, a field you wished an entry had. A person reads it. For a word with no entry use report_missing_word instead. Name the word and the language when the note is about one entry.

The immit dictionary entries that match a string: a headword prefix, an alternative spelling, an inflected form (食べた finds 食べる, walked finds walk) or, for Japanese, a kana reading (たべる finds 食べる). Results are ranked: an exact headword or spelling first, then an exact inflected form or reading, then the prefix matches, each rank by frequency with the most common first. Each result is a whole entry, as lookup_word returns it. Use it to turn an inflected form, a reading or a fragment into the headword, or to list the words that start with a string; use lookup_word when you already hold the dictionary form.