MCP App Store
by KeenEthics
LangLearn icon

LangLearn MCP Server for ChatGPT

By CTO Teachings LLC

Overview

Use your LangLearn vocabulary to create stories, conversations, and audio lessons, then save them to your LangLearn library. Browse your existing passages and retrieve the word lists and learning rules for your selected language. Requires a LangLearn account with a learning language configured. This connection provides learner tools; administrative language and course authoring are unavailable. Generated content is validated by LangLearn and can be rejected when it does not meet the learning rules.

How to connect

  1. ChatGPT: Open the LangLearn listing in the ChatGPT app directory and choose Connect. ChatGPT listing LangLearn exposes 13 tools here.

Tools

learner_get_allowed_vocabulary

ChatGPT
Page through a saved generation request's allowed wordforms, up to 500 per page.

learner_get_generation_definition

ChatGPT
Read the current requirement fields, defaults and validation guidance before the first Save. Creates no request.

learner_get_generation_item_requirements

ChatGPT
Read one saved story or conversation plan using requirementsId and storyNumber. The server resolves revision. Never redraws assignments.

learner_get_generation_requirements

ChatGPT
Read a saved generation request and its writing instructions. Does not create or change it.

learner_get_generation_status

ChatGPT
Check drafts and audio readiness without resending content. Resumes interrupted publication after its lease expires.

learner_get_learning_settings

ChatGPT
Read your first language, learning languages and learning settings. Reading changes no settings or word statuses.

learner_get_passages

ChatGPT
Everything already in my library, newest first: STORIES, CONVERSATIONS (stories with two named speakers) AND audio LESSONS (lessons still rendering are skipped). Read them to avoid repeating a subject or to continue a plot. On a first run this is simply empty. PAGED: pageSize (1-50, default 10). Pass the returned nextStartAfter back as startAfter and read hasMore rather than guessing from a short page. You rarely need more than the first page or two — the newest stories are the ones a new story is most likely to repeat.

learner_get_wordform_lists

ChatGPT
THE VOCABULARY A STORY MAY USE. My lists are WORDFORM lists — a frequency list of written forms, so "рік", "роки" and "років" are three separate entries with three ranks, and an inflection you do not find on a list is NOT covered by its dictionary headword. WRITING FOR A SAVED REQUEST (my browser prompt gave you a requirementsId): ALWAYS pass that requirementsId here FIRST. I then answer with exactly the words THAT request lets you use: requestWords[] — range mode: the wordforms ranked minWordRank..maxWordRank that I asked to be taught, the words to build the story around (pass startAfterRank 0 to page the more common wordforms below minWordRank, which you may also use); requiredWords mode: every wordform at or below maxWordRank, plus requiredWords[], the words the story must use. Also EVERY rule the saved request sets: minWordRank, maxWordRank, newWordLine (the range words above it are the new ones), oneNewWordPerSentence (zero or one new word per sentence — sentences with no new word are welcome), the rangeWordsPercent minimum, firstLessonsWordsKnown, and the length band. Take the words from here BEFORE writing — never write first and wait for learner_upload_story to name words. WITHOUT a requirementsId: known[] (paged, most-used first) and newWords[], the ~10-wordform window above knownRank, offered as a starting point for a plain story. Paging: pass nextStartAfterRank back as startAfterRank and read hasMore rather than guessing from a short page. Also here, so you need no second call before writing: the length band, suggestedNames (the language's 10 names and 10 places), and rules[] — exactly what learner_upload_story will check. There is no limit on words per sentence.

learner_get_words

ChatGPT
Browse explicit Known, Unknown or ranked vocabulary with paging and independent exclusions. Translate in the requested language, otherwise the saved first language.

learner_save_generation_requirements

ChatGPT
Save story, conversation or lesson choices with a retry key. Returns the saved ID and writing instructions.

learner_save_generation_requirements_v2

ChatGPT
Save current numbered story/conversation plans with separate per-pool quotas. Replaces old saved requirements on edits. Use item Get before each story.

learner_submit_lesson

ChatGPT
Submit an audio lesson for saved requirements. Clean content queues audio; rejected content stays private for repair. Keep submissionId on retries.

learner_submit_story

ChatGPT
Submit a story or conversation for saved requirements. Clean content publishes; rejected content stays private for repair. Keep submissionId on retries.

learner_update_generation

ChatGPT
Repair a rejected draft using exactly one nonempty story, conversation or lesson section matching its saved kind. Retain patchKey on retries.

learner_upload_lesson

ChatGPT
Validate and publish ONE audio lesson. A human is going to listen to it, so write something that actually teaches: a scene, a few useful phrases, and the learner speaking out loud. You write the WRITE A CONVERSATION, NOT A BARE LIST OF WORDS. Two people saying things to each other for a reason, where each line follows from the one before. The rules below name words a lesson owes — but a lesson that flows and owes one word less is better than one that covers everything and reads like errands, and one sentence may satisfy several rules at once. Compose the conversation first, then check what it is still missing. Before you send it, read it back: if a line exists only to deliver a word, rewrite it or cut it. CONTENT; the server builds the audio script from it and checks everything before any audio is made. Retries with the same uploadId are idempotent. A rejection names exactly what to fix — fix it and send the lesson again. Read every response: one may ask you to do something first. DRY RUN FIRST. Send dryRun: true to run every check without storing anything; the answer is ok plus findings. Fix every finding and repeat until ok is true, then send the same lesson without dryRun. Before that, have a validation subagent check that every line is natural. THE CONTENT (lesson): sentences — the conversation, in the language being learned, no word twice in one sentence. How many is the BAND's: up to 12, and from the middle of a course a lesson runs 9 to 12. Write the number the conversation needs, not the minimum — the extra sentences exist so the words you owe have somewhere natural to go. hookLines — 2 to 4 lines of the mini-conversation the lesson opens and closes on, each made of whole sentences, and no line starting with all the words of the line before it (merge such lines into one); wordformsToLearn — the wordforms the lesson teaches, one entry each as {wordform, translation}, every wordform ONE word said whole by a sentence and spelled exactly as that sentence spells it, and translation its meaning in my first language; reviewWordforms — earlier words to ask for again; syllables — for every word of every taught wordform, {"parts": [...]} whose parts join back into the word (English also needs phonemes, one IPA entry per part); meanings — my first-language meaning of every review wordform and whole sentence (a taught wordform's meaning rides on its own entry, not here); situations — each sentence's one-line scene ("You bump into a friend in the street."); cues — for each sentence, the question that asks for it ("How do you say: hi?"); rolePlay — 2 to 4 {cue, sentence} scene cues ("A friend waves at you. Say hi."); framing — hookIntro ("Listen. By the end of this lesson you'll be able to say this."), sentenceIntro ("Listen to this sentence."), meaningIntro ("It means:"), walkIntro ("Listen and repeat."; it may use {sentence} for the sentence number and {index} for the word the build-up starts at, e.g. "Sentence {sentence}. Listen and repeat.", and never a digit), learnIntro ("Let us learn how to say '{meaning}'" — it must carry {meaning} exactly once), syllableIntro ("Repeat these syllables"), repeatIntro ("Repeat after me" — opens the drill of a one-syllable word), meaningQuestion ("What does {word} mean?" — it must carry {word} exactly once), reviewIntro and dialogueAgain ("Listen to the conversation again."), all in my first language; optionally introProse and names. WHAT THE SERVER BUILDS: the hook; for each sentence with a new wordform, its scene, the sentence, "It means: …", "Listen and repeat.", a syllable drill for each new word of two or more syllables, chunks built from the END (each said twice, the second slowed), and a cue asking for it, then the cue of the previous sentence for you to say again; then "what does it mean?" for every new word and every sentence; then a review that asks for every word and every sentence once and plays the role-play cues; the hook again. Keep meanings, scenes and cues short. THE BANDS. A course lesson gets its band from …

learner_upload_story

ChatGPT
Save a STORY — or a CONVERSATION, a story with two named speakers — that you wrote for me. It is CHECKED and rejected if it breaks the rules below — read them before writing, not after a rejection. BOTH TITLES: send title in toLearnLang and firstLangTitle in firstLang. Neither may be blank. Send a unique uploadId for this story and keep it EXACTLY the same on transport/TTS retries. A changed story needs a NEW uploadId. The server publishes both language versions together, or neither: both wordform lists and both audio clips must succeed before either becomes visible. ONE STORY, TWO CATALOGUES, AND THEY ARE JUDGED DIFFERENTLY. You send each sentence in BOTH languages. The toLearnLang half is filed as my story and is checked 100% against the word rules below. The firstLang half is NOT checked against anybody's level: it is re-tokenized and RATED, then filed as a real story in the firstLang catalogue for people learning THAT language from mine. A hard firstLang sentence is never a rejection — it just rates that story higher and serves a more advanced reader. So write the firstLang half as natural prose, not as a word-for-word crib. THE SHAPE. A story is a LIST OF SENTENCES, not a blob of text. Each entry carries one toLearnText and the same sentence as firstLangText — I read them stacked, one under the other, so a firstLang line covering two sentences at once lands under the wrong one. BOTH LANGUAGES ARE NAMED, ALWAYS. toLearnLang and firstLang are both REQUIRED and neither is guessed from my settings — an upload files a story in EACH of them, so each one has to be chosen rather than inferred. ONE KIND OF STORY, IN ONE OF TWO MODES. Every upload sends maxWordRank (with requirementsId it must EQUAL the saved request's Max Word Rank; otherwise use what I asked for or learner_get_wordform_lists.defaultMaxWordRank). With a requirementsId, FIRST call learner_get_wordform_lists with that same requirementsId: its requestWords are the words I asked for, and you write from them. EVERY RULE BELOW IS THE SAVED REQUEST'S — read its values from learner_get_wordform_lists (oneNewWordPerSentence, the rangeWordsPercent minimum, firstLessonsWordsKnown, minWords/maxWords). With a requirementsId, leave out minWordRank, minWords, maxWords and requiredWords — they are the saved request's; a different value is refused. Then exactly one mode: RANGE (the default): the range words (minWordRank..maxWordRank) are the NEW words; every word below minWordRank is free to use. Every word is ranked at or below Max. Meet or exceed the saved minimum % of story words that are range words (default 7%; 20% is only the highest configurable minimum, not an actual-content ceiling): validation credits each DIFFERENT range wordform once, divided by ordinary ranked story-word occurrences. Declared names/places do not count; repeating one word earns no extra credit. You pick WHICH range words to use — you need not use them all. A rejection says exactly how many different range wordforms to add. Without requirementsId send BOTH minWordRank and maxWordRank, each at least 200, at most 50 apart. REQUIRED WORDS: only the required words are new; every other word is ranked at or below maxWordRank. Every listed word appears (with oneNewWordPerSentence: at most one per sentence). With requirementsId the list is the saved request's; outside the browser send requiredWords (1-10 words, never together with minWordRank). NAMES AND PLACES. Use the language's names and places: learner_get_wordform_lists.suggestedNames (10 names + 10 places), as many as the story needs — there is no cap. Mark a person's name with kind "name" and a place with kind "place" in toLearnTokens. A person must be one of those names. A declared name/place never sets the story's level and never counts toward the range-word percentage; a declared word that is NOT one of them still obeys maxWordRank. CONVERSATIONS. When the saved request (requirementsId) is a CONVERSATION, the story is two people talking: send speakers — EXACTLY two, each {toLearnName, firstLangName} — and give EVERY sen…

Capabilities

Writes

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/langlearn"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/langlearn)