MCP App Store
by KeenEthics
Tokki Talk: Learn Korean icon

Tokki Talk: Learn Korean MCP Server for ChatGPT

By Little Oboe LLC

Overview

Tokki Talk is a Korean tutor companion and study helper for speaking, listening, reading, and vocabulary. Combine personalized Korean audio lessons, AI speaking practice, contextual explanations, flashcards, and spaced repetition using your own learning material. Explore words and expressions in context, Korean grammar, nuances between similar phrases, Hangul reading, and natural usage. ChatGPT explains and conducts chat practice; Tokki Talk brings your saved vocabulary, lists, review schedule, and interactive learning views into the conversation. Use Tokki Talk alongside your Korean classes, textbooks, and study resources. Bring notes or short excerpts you are allowed to use, Korean text, a screenshot, subtitles, a menu, a TOPIK practice passage, K-drama dialogue, or a short K-pop lyric excerpt. Ask for useful words and reusable expressions with meanings and examples. When you ask to save them, Tokki Talk keeps them in your account or a named list. Check which items are already saved or scheduled for review. Practice Korean conversation with roleplays using your own lists, and ask for corrections after each turn. Ask what is due today for a quiz drawn from your actual spaced repetition (SRS) queue. Supported learning views include word cards, sentence breakdowns, examples, corrections, and list summaries; speaker controls can play Korean pronunciation subject to your account's voice permissions. A mention in chat does not automatically earn review credit. Continue speaking, listening, and reading in the companion Tokki Talk app: AI voice conversations and audio lessons with high-quality Korean voices, podcast listening with transcripts where available, stories with transcripts, and flashcard practice. Follow Korean YouTube subtitles with Tokki Watch, inspect words in context, and save useful vocabulary while studying. YouTube subtitle following runs in the companion study surfaces; in ChatGPT, bring subtitle text or screenshots for explanation and saving. Practice natural Korean speaking with AI voice conversations, rather than a human native-speaker matching service. Its import flows accept Anki files and learner-provided Quizlet card text or an explicit Safari visible-card export. This connector uses your Tokki Talk study state; it does not directly access your Anki or Quizlet account or run the app's live voice conversation inside ChatGPT. The connector also supports previewing audio lessons and curricula from saved lists, then requesting generation after you confirm the preview, subject to account permissions and quotas. Open generated lessons in Tokki Talk for playback. Connect your Tokki Talk account for saved state and persistent actions; ordinary Korean explanations can be answered directly without reading your account.

How to connect

  1. ChatGPT: Open the Tokki Talk: Learn Korean listing in the ChatGPT app directory and choose Connect. ChatGPT listing Tokki Talk: Learn Korean exposes 28 tools here.

Tools

approve_curriculum

ChatGPT
Approve a Korean course: accepts a 'proposed' plan (making its lessons generatable) or activates a pending revision created by revise_curriculum. Call only after the user has seen the plan from get_curriculum / revise_curriculum and agreed. Approval itself generates nothing; follow with generate_curriculum_lessons.

create_curriculum

ChatGPT
Confirm a previewed Korean course and start its planning job. This is PAID and runs for several minutes in the background: call it only after the user approved the words and lesson count from preview_curriculum, passing the same arguments plus its preview_token (refused with curriculum_preview_stale if the list changed). Returns plan_id with curriculum_status 'planning'. Then poll get_curriculum; when it is 'proposed', present the lesson plan for review (revise_curriculum / approve_curriculum). A list can have only one active course.

create_saved_list

ChatGPT
Create an empty named Korean vocabulary list and return its list_id, name, clickable Tokki Talk deep_link, and mutation result. name is required. Supply list_id only when a stable id is needed for idempotent creation; the server chooses one when it is omitted. Supply mutation_id to make a retry idempotent. Successful list results include deep_link. After creating, populating, renaming, or changing a list, offer that URL once as an 'Open in Tokki Talk (app or web)' Markdown link. Do not offer a link for an error or repeat it during passive reads unless the user asks to open the list.

discard_curriculum

ChatGPT
Remove a Korean course from the learner's library, e.g. to re-plan it with different words or guidance (a list can have only one active course). Allowed for proposed, ready, complete, or failed courses; refused while planning is running or a lesson is generating. Already generated lessons and earned review credit are kept. Confirm with the user first.

generate_curriculum_lessons

ChatGPT
Generate the next Korean audio lessons of an approved course (PAID, minutes per lesson, background). Prepares up to max_lessons (1-10, default 1) contiguous not-yet-generated slots atomically; connected-story courses prepare one at a time. Returns the pending lesson ids. Poll with list_lessons (plan_id) or get_curriculum.

generate_lesson

ChatGPT
Confirm a previewed selection and enqueue the Korean audio lesson. This is a PAID, minutes-long background job: call it only after the user approved the words shown by preview_lesson. Pass the same list, selected_words, target_duration_minutes, items_count and include_known as that preview plus its preview_token; the server recomputes the preview and rejects with lesson_preview_stale (returning a fresh preview) if anything changed. Returns the pending lesson_id, a deep_link, and next_step. One lesson generates at a time per learner. Poll with list_lessons.

get_curriculum

ChatGPT
Read one Korean curriculum: curriculum_status (planning, proposed, ready, complete, failed), each planned lesson's theme, objective, new_words, review words, editability and generation progress, the current revision number, and a next_step hint. Give plan_id, or list_name/list_id for that list's current course; with no arguments it lists all of the learner's courses. Use it to poll after create_curriculum and to show the plan before approval.

get_due_words

ChatGPT
Return the learner's current Tokki Talk due review queue of Korean words, capped by the learner's daily new-card budget. Use when asked what to review or to quiz the learner on due items. Each result has exact study text, kind, gloss, due time, and SRS status where available. This is a review selection, not the full due backlog; use get_learner_snapshot for the backlog count. Use list_words_in_saved_list for a named list.

get_learner_snapshot

ChatGPT
Return a compact Tokki Talk study-state snapshot: encountered-vocabulary count, due review count, weak-spot items, and saved-list names. Encountered does not mean mastered. Use when the learner asks about their Tokki Talk progress, review, or lists; ordinary Korean explanations do not need Tokki Talk lookup.

get_word_origin

ChatGPT
Return when, how, and why a Korean vocabulary word was added. Identify it by canonical_form or a previously returned item_id. The result includes the add timestamp, source, saved reason, any originating chat-session title and message excerpt, and a clickable Tokki Talk deep_link. found=false means the word is not tracked; has_origin=false means it is tracked but has no recorded add origin.

get_words

ChatGPT
Look up specific words in the learner's Tokki Talk state and return their saved gloss, SRS schedule, note, saved-list membership, and clickable Tokki Talk deep_link. Use only when learner context matters, such as whether a word is tracked or saved; answer ordinary Korean meaning questions directly. Identify words by canonical_forms (dictionary forms) and/or previously returned item_ids. Existing words without an SRS card have srs=null; untracked words are returned in not_found. A saved or scheduled word is not proof that the learner knows it. Set include_sentences=true to include up to three recent saved example sentences per word. Set include_dictionary=true only when expanding a Tokki Talk word card to read additional cached/corpus definitions, examples, and reviewed synonyms or antonyms for that exact identity. This optional read does not generate dictionary content, save words, or award review credit. Missing dictionary content returns dictionary=null.

import_story_reading

ChatGPT
Append approved Korean source text as a new reading in one existing ready imported Story collection. title and text are required; source text is tokenized and stored without rewriting. Supply the preferred series_id, or exact collection_title when the id is unavailable. The tool never creates a collection. Returns the new episode_id, canonical_text, artifact hash, and a custom-scheme Tokki Talk deep_link because Story collections do not yet have a public HTTPS app-open route.

list_lessons

ChatGPT
Return the learner's recent Korean audio lessons with status (pending, ready, failed, cancelled, consumed, expired), generation_phase, failure_reason, source list, target words, and deep_link. Filter by lesson_id, list_name/list_id, or curriculum plan_id. Use it to report progress after generate_lesson or generate_curriculum_lessons.

list_saved_sentences

ChatGPT
Return saved sentences (example sentences and phrases), including sentence_id, headword, Korean text, English translation, and mark. Optionally filter by canonical headword or cap the result with limit.

list_words_in_saved_list

ChatGPT
Return the Korean words in one vocabulary list, including each word's dictionary form, gloss, part of speech, SRS state, and clickable Tokki Talk deep_link. Resolve the list by case-insensitive list_name or list_id. If it is absent, returns found=false and available list names. Use search_term to match any part of a dictionary form, including part of a multi-word phrase; omit it for an unfiltered page. Use semantic_query for topic/meaning search, or saved_filter_id for an existing personal filter. Combine with lifecycle and POS restrictions; this never creates a chip or changes list membership/SRS. Pending/partial semantic results include evaluation_id; resume with that ID without scheduling paid work. Results are bounded to 100 per page; pass the returned result_snapshot_fingerprint as snapshot_fingerprint when requesting next_offset. Set include_sentence_coverage=true when checking which returned members lack saved sentences; it returns authoritative per-member counts and an uncovered-member summary. Successful list results include deep_link. After creating, populating, renaming, or changing a list, offer that URL once as an 'Open in Tokki Talk (app or web)' Markdown link. Do not offer a link for an error or repeat it during passive reads unless the user asks to open the list.

play_pronunciation

ChatGPT
Play Korean pronunciation on an explicit speaker-button click. Uses Tokki voice preferences; never records a review.

populate_saved_list

ChatGPT
Create or reuse a named vocabulary list, or populate an accessible existing list_id, with the supplied Korean study forms and optional example sentences. Each study_form is the exact form to study and must not be replaced with a lemma. Supply baseline_glosses_en for English definitions to save on each list item; gloss is only a sense-disambiguation hint and is not saved as its definition. Items and their optional sentences are processed independently: one failure does not roll back successful siblings. Returns an ordered result for every input, outcome success, partial_success, or failed, the authoritative final list contents, and clickable Tokki Talk deep_links for the list and each resolved item. mutation_id makes an exact retry return its stored result; reusing it with different content is rejected. Successful list results include deep_link. After creating, populating, renaming, or changing a list, offer that URL once as an 'Open in Tokki Talk (app or web)' Markdown link. Do not offer a link for an error or repeat it during passive reads unless the user asks to open the list.

preview_curriculum

ChatGPT
Free preview of a Korean curriculum (a finite, ordered course of lessons) built from one saved list the learner owns: the exact words in teaching order, the derived lesson_count, per-lesson capacity, whether the list already has an active course, and a preview_token. Nothing is created. ALWAYS call this first and confirm the words and lesson count with the user before create_curriculum. A course covers at most 50 words; use selected_words (exact forms, in order) for a subset or a custom order. Lesson count is derived from the word count and target_duration_minutes (15-30), not chosen directly.

preview_lesson

ChatGPT
Free, deterministic preview of the Korean words an audio lesson generated from one of the learner's saved lists would teach. No LLM call, nothing is created. Returns selected_words (fitted to target_duration_minutes), deferred_words that did not fit, review words, grammar focus, the available_words pool with each word's state, and a preview_token. ALWAYS call this first and show the user the words before generate_lesson. To change the selection, call again with selected_words (exact forms from available_words in any state: 'new', 'learning', or 'known').

remove_from_saved_list

ChatGPT
Remove one or more exact displayed Korean study_forms from an existing vocabulary list and return the mutation result. Resolve the list by case-insensitive list_name or list_id; missing lists are rejected and are never created. item_id may disambiguate exactly one study_forms entry. Forms not already in the list are reported as not_present rather than errors. Supply mutation_id for an idempotent retry. Successful list results include deep_link. After creating, populating, renaming, or changing a list, offer that URL once as an 'Open in Tokki Talk (app or web)' Markdown link. Do not offer a link for an error or repeat it during passive reads unless the user asks to open the list.

rename_saved_list

ChatGPT
Rename an existing Korean vocabulary list and return the mutation result. Resolve it by its current case-insensitive list_name or list_id, set new_name, and never create a missing list. Built-in system lists cannot be renamed. Supply mutation_id for an idempotent retry. Successful list results include deep_link. After creating, populating, renaming, or changing a list, offer that URL once as an 'Open in Tokki Talk (app or web)' Markdown link. Do not offer a link for an error or repeat it during passive reads unless the user asks to open the list.

render_learning_view

ChatGPT
Render a compact Tokki learning view after preparing task-specific data. Use for word cards, vocabulary groups or selection, sentence breakdowns, examples or corrections, list summaries, due review, or roleplay setup. In ChatGPT, call this after creating or updating a vocabulary list to show list_summary with its returned list ID/name and actual items; writes do not display UI automatically. When the host renders this component, it is the presentation: do not repeat its words, definitions, examples, translations, or status badges in prose or Markdown. After rendering, use at most one short acknowledgement or helpful next step. Keep item_id, study_item_id, homograph_id and raw POS codes internal; never print a verified-identity block. Only provide a text fallback if the host cannot display the component. For sentence_breakdown prepare nonempty items yourself, with dictionary study_form, original display_form, meaning and grammar detail. Preserve original sentence spacing in source_text and source_tokens; token text concatenates verbatim. Supply item_id from get_words when available. POS, SRS labels and counts come from Tokki lookup results; omit them when unknown. This tool only displays supplied data and never saves it. The text result remains useful in clients without UI.

revise_curriculum

ChatGPT
Check a proposed source assignment against the approved course. Moving or reordering words currently requires discarding and creating a new audited course because each lesson's objective, scene, and sense anchor are tied to its original words. This tool rejects changed assignments; unchanged slot edits can still create a revision under the normal approval fence.

save_sentence

ChatGPT
Save one or more Korean example sentences or phrases and return saved sentence ids and rejected_items. Prefer the sentences batch array; its entries use headword, text_target, text_en, and optional mark. For a direct MCP call, provide headword for each sentence plus its Korean text and English translation. The top-level single sentence fields remain a fallback. Supply mutation_id for an idempotent retry.

save_vocab

ChatGPT
Save one or more Korean vocabulary items and return saved, already_present, and rejected_items. Each study_form must be the exact requested Korean form; a multi-word phrase or expression is one item, not separate words. The server resolves citation form and lexical identity. list_name or list_id adds items only to an existing list and also enrolls them for review; it never creates a list. mark_seen defaults to true and records items as encountered for review; set it false only for unscheduled, aspirational list building. With no list and mark_seen=false, items are saved without a review card. reason is stored as add provenance. Supply mutation_id for an idempotent retry. Successful list results include deep_link. After creating, populating, renaming, or changing a list, offer that URL once as an 'Open in Tokki Talk (app or web)' Markdown link. Do not offer a link for an error or repeat it during passive reads unless the user asks to open the list.

set_vocab_note

ChatGPT
Set or clear the freeform study note for a Korean vocabulary item and return the mutation result. Provide the required study_form and use item_id when known to identify the item exactly. note overwrites the prior note; an empty string clears it. Supply mutation_id for an idempotent retry.

update_saved_sentence

ChatGPT
Update an existing saved sentence in place and return its sentence_id and mutation result. Supply sentence_id plus at least one of text_target (the Korean text), text_en, mark, or headword; only supplied fields change, while the original save time is preserved. Use list_saved_sentences to obtain a sentence_id. Supply mutation_id for an idempotent retry.

update_word_status

ChatGPT
Change an owned word's familiarity on an explicit SRS-picker click, or undo that exact change. Never grade a review.

Capabilities

InteractiveWrites

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/tokki-talk-learn-korean"><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/tokki-talk-learn-korean)