Konakai tool reference

The Konakai plugin is one MCP server with three tools. The AI does the teaching; the tools read and save the learner’s study. How to add Konakai.

Connecting

Server: https://api.konakai.ai/mcp, Streamable HTTP.
Sign-in: OAuth 2.1 authorization code with PKCE, on konakai.ai. Scopes: learning:read, learning:write and offline_access (refresh tokens). Each host uses its own pre-registered client; dynamic client registration is off.
Account: a Konakai account, for adults 18 and over in the United States.
Results: JSON in a text block, also as structuredContent.

`start_here`

Open Saved Study · Read · scope learning:read

Changes nothing. Counts against the daily allowance.

Inputs
Field	Type	Required	Limits	Meaning
`target_language`	string	no	matches ^[a-z]{2,3}$	Code of the language the learner is studying, when they have named it.

Returns: target_language, native_language, material_languages, due_cards, open_mistakes, recent_lines, recent_topics, story, preferences, instruction. Empty fields are left out.

`get_material`

Get Saved Material · Read · scope learning:read

Changes nothing. Counts against the daily allowance.

Inputs
Field	Type	Required	Limits	Meaning
`target_language`	string	no	matches ^[a-z]{2,3}$	Code of the language the material is in: the one this chat studies.
`categories`	list of string	yes	each one of "lines", "words", "mistakes", "topics", "story"; at least 1 item; at most 5 items	What to get: saved lines, words, mistakes, topics, or the story.
`search`	list of string	no	at least 1 item; at most 10 items; each at least 1 character; each at most 100 characters	Terms to search for, each on its own, in either language. Omit to list without searching.
`due_only`	boolean	no	—	Only lines and words due for review.
`resolved_mistakes`	boolean	no	—	Mistakes the learner has since gotten right, instead of open ones.
`page`	integer	no	from 1; to 100	Which page of each listing, from 1.

Returns: lines, words, mistakes, topics, more, story, searches, instruction. Empty fields are left out.

`save_material`

Save Study · Write · scope learning:write

Adds lines, words, mistakes and topics to the learner’s study; a rating schedules that card’s next review; a corrected mistake is marked resolved. The story and preferences it sends replace the saved ones, and an empty preferences value clears them. Nothing already saved is saved twice. It sends nothing to anyone and takes no payment.

Inputs
Field	Type	Required	Limits	Meaning
`target_language`	string	yes	matches ^[a-z]{2,3}$	Code of the language being learned.
`native_language`	string	yes	matches ^[a-z]{2,3}$	Code of the learner's native language, the one the translations are in.
`items`	list of object	no	at most 60 items	Every target-language line you taught or had the learner practice since your last save in this chat, in its correct form, at the sentence, phrase and collocation level, including lines they asked for, and each word you taught on its own. Don't re-list a word that appears inside a listed line; list a word on its own only when no line contains it.
`items[].content_type`	string	yes	one of "word", "collocation", "phrase", "sentence"	What this line is: "word" = one atomic target-language word (exactly one content_array element), a one-word sentence included; "collocation" = a fixed multi-word combination; "phrase" = a short multi-word unit below clause level; "sentence" = a clause or sentence of two or more words. Everything except "word" has two or more content_array elements.
`items[].content`	string	yes	at most 1000 characters	Full target-language display string.
`items[].romanization`	string	no	at most 1000 characters	Canonical written romanization/transliteration of the full non-Roman target-language display string; never an invented sound-it-out respelling.
`items[].translation`	string	yes	at most 1000 characters	Full-line translation in learner's native language.
`items[].content_array`	list of string	yes	at least 1 item; at most 100 items; each at most 500 characters	Ordered target-language vocabulary elements modeled in this line, one element per array entry. Each element is one atomic word of the target language — never a clause, sentence, multi-word combination, or standalone punctuation; decompose multi-word lines into their word elements and omit punctuation. Word boundaries are the target language's, not the learner's: a single word in the learner's language often glosses several target-language words, and each of those is its own element.
`items[].romanization_array`	list of string	no	at most 100 items; each at most 500 characters	Canonical written romanization/transliteration elements aligned by position with content_array, one per array entry; never invented sound-it-out respellings.
`items[].translation_array`	list of string	no	at most 100 items; each at most 500 characters	Translation/gloss elements aligned by position with content_array, one per array entry. Required for multi-element content_array; give each vocabulary element its own single learner-language gloss. Each gloss is one rendering — never slash-joined or comma-joined alternates within one entry.
`mistakes`	list of object	no	at most 30 items	The learner's mistakes since your last save in this chat. A mistake is an answer the learner gave to a prompt with an expected answer that was wrong and stayed corrected. An answer the learner says was misheard is what they say they said, and no mistake once it is accepted.
`mistakes[].expected`	object	yes	—	The right answer, as a card.
`mistakes[].expected.content`	string	yes	at most 1000 characters	The target-language line.
`mistakes[].expected.romanization`	string	no	at most 1000 characters	Canonical written romanization of `content`, for a non-Roman target language.
`mistakes[].expected.translation`	string	yes	at most 1000 characters	The line's meaning in the learner's native language.
`mistakes[].answer`	object	yes	—	What the learner answered, as a card: an answer they said in the target language is its `content`, with `translation` a literal rendering of those words, never the meaning they were aiming for; a meaning they gave is its `translation`, with `content` that meaning put into the target language: a different line from the one they were asked about.
`mistakes[].answer.content`	string	yes	at most 1000 characters	The target-language line.
`mistakes[].answer.romanization`	string	no	at most 1000 characters	Canonical written romanization of `content`, for a non-Roman target language.
`mistakes[].answer.translation`	string	yes	at most 1000 characters	The line's meaning in the learner's native language.
`mistakes[].direction`	string	yes	one of "production", "reception"	production when they said or wrote the right answer wrong; reception when they misunderstood a target-language line.
`ratings`	list of object	no	at most 60 items	One rating for each saved line or word you quizzed the learner on since your last save in this chat.
`ratings[].text`	string	yes	at least 1 character; at most 1000 characters	The saved line or word you quizzed, as written.
`ratings[].rating`	number	yes	one of 1, 2, 3, 4	Your rating of the learner's recall: 1 = wrong or forgotten, 2 = mostly wrong, 3 = mostly right with a minor form-only issue (meaning intact), 4 = correct, near-perfect, or confidently known.
`mistakes_corrected`	list of string	no	at most 60 items; each at least 1 character; each at most 1000 characters	The right answer of each open mistake the learner answered correctly since your last save in this chat. A record is answered correctly when the learner's first attempt at it, before any hint, is right; a turn the tutor said it did not catch is no attempt.
`topics`	list of string	no	at most 30 items; each at least 1 character; each at most 200 characters	Each situation, theme, grammar point or pronunciation point the learner studied since your last save in this chat, one short line each, the situations inside the story included.
`story`	string	no	at least 1 character; at most 1000 characters	The story as it now stands, replacing the saved one: rewritten from the saved story and this chat when the chat continued it, or the new story when the learner asked to start fresh. Keep only what the story needs to continue: where the learner is and who is with them, what they are after, open threads, the places and situations it has already used, briefly, so it moves on to new ones, and what happens next. Drop whatever no longer matters.
`preferences`	string	no	at most 1000 characters	How the learner has asked to be taught from now on, as a short list: the saved preferences with what they asked for in this chat added or changed, replacing them. Only what they asked for; name the language when one holds for a single language. An empty string clears them.

Returns: saved, already_saved, rejected, unmatched, instruction. Empty fields are left out.

Errors

An invalid entry in save_material: the rest is saved, and each invalid one comes back in rejected with its problem and a suggestion. A rating or correction that matches nothing saved comes back in unmatched.
Invalid arguments: a tool error that names the problem.
The daily allowance used up: a tool error of type usage_limit; it resets the next day.
No sign-in, or an expired or revoked token: HTTP 401 with a WWW-Authenticate header pointing to the protected-resource metadata.
A missing scope: a tool error that asks for the scope.
Too many requests: HTTP 429 with Retry-After.

Limits

25 tool calls a day on the free plan, 100 on a paid plan. A typical chat uses three to six.
300 requests a minute per account, counting every MCP message.

Data and support

Konakai keeps only what save_material saves, in the learner’s account. The Privacy Policy says how it’s used and how to delete it. Support: [email protected].