Save vocabulary from any AI chat; see each word in every language you know; review it with FSRS.
Scored across 5 tools
The five tools cover distinct operations: reading languages, initial language setup, looking up a word, saving one word, and saving multiple words. The boundary between save_word and save_words is mostly clear from the descriptions, but a confusing mention of a 'save_languages' parameter in save_word slightly muddies the distinction.
All tool names use a consistent snake_case verb_noun pattern (get_languages, lookup_word, save_word, save_words, set_languages). The singular/plural distinction between save_word and save_words is intuitive and predictable.
Five tools is a well-scoped set for a vocabulary-capture connector, with each tool earning its place: setup, language retrieval, lookup, and single/batch saving. No obvious redundancy or missing tool bloat.
The surface covers account setup, language retrieval, lookup, and both single and batch saving, which handles the core capture workflow. However, there is no tool to delete or manage saved vocabulary (e.g., remove a word or list saved words), leaving minor gaps for full lifecycle management.
Returns the languages this Koboshi account is set to: the language meanings are written in, the language being learned, and any others the user has added. The same list arrives with the connect-time instructions and on the [account: ...] tag of every result, so this tool is for the case where that list was missing or has just been changed. The account setting is the only place these languages live; they are unrelated to the languages on the user's device or in this conversation. Takes no arguments and writes nothing.
Inputs: no named parameters.
Returns what Koboshi holds for a word, and the same word in every language on the account (the result names them). Koboshi's own card is used where it has one. Where it has none, the result carries Koboshi's rules for the word's card and for its card in each missing language; a second lookup carrying them in "card" and "translations" returns them checked, in one list, marked as written by the AI in the chat. Reads only: nothing is written or saved. save_word performs this lookup itself, so this tool is for looking a word up without keeping it. learning_language is the language the word is written in.
Inputs: word (required), card, detail, translations, +2 more.
Adds one word, phrase or idiom to the user's own Koboshi vocabulary, so their spaced-repetition reviews bring it back. The word is saved in the one language it is written in. Koboshi looks it up as part of this call: when it already holds a card the save completes, and when it is new the result carries the card rules and the call is repeated with the finished card. Saving the same word twice does not create a second entry. One call saves the one word it was given and nothing else, unless "save_languages" names the words the user picked. Several different words from one request belong in save_words instead. A saved phrase or idiom counts one word for every 10 words it holds toward the account's 100-word daily limit, so a long one can use up more than one. The result shows the word in every language on the account, for reading. Cards Koboshi lacks are written by the AI in the chat from Koboshi's rules and sent in "translations"; Koboshi checks them and keeps them for its own review, and they add nothing to the vocabulary by themselves. Optional: "sense" names which meaning of an ambiguous word was wanted: a new word is saved with it, and a word already saved adds that meaning (its first meaning stays); "context_sentence" keeps the sentence the word appeared in as the user's own private example.
Inputs: word (required), card, sense, concept, +5 more.
Adds several DIFFERENT words, phrases or idioms to the user's own Koboshi vocabulary in one request, for example a set of words from earlier in this conversation, sent together rather than one at a time. One word belongs in save_word, including a word wanted in more than one language. Each entry carries the language that word is written in. Words already in the vocabulary are reported as skipped; words new to Koboshi come back listed as needing a card, and cards sent with the first call are used straight away. Up to 50 items a call; a long phrase counts more than one of the account's 100-word daily limit, one word for every 10 words it holds. This connector can see only the current conversation.
Inputs: words (required), cards.
First-time setup for a Koboshi account that has no languages yet: the language meanings are written in, the language being learned, up to four more the user knows or is also learning, and the month and year the user was born, which Koboshi asks every new account, as its website does. It applies to an account still waiting to be set up. An account whose languages are already chosen keeps them, and the result says what they are; a change to them happens in the Koboshi app, under Settings.
Inputs: primary_language (required), learning_language (required), birth_year, birth_month, +2 more.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Save words from your AI chats as spaced-repetition flashcards, with meanings, IPA and audio.
Free: turn your AI chats into spaced-repetition vocabulary. 13 tools, reads and writes.
Save words to your Polidict vocabulary and practice them with spaced repetition from any AI chat.
Spaced-repetition flashcards your AI writes, quizzes you on by voice, and schedules with FSRS.
Add one secure layer between your agents and this server.
