Yomi — Data & Privacy

Log in to LINE via the passwordless secondary-device flow. Requires 設定 > 我的帳號 > 允許自其他裝置登入 enabled on the primary phone — with it off LINE never prompts that phone and NO login can succeed, so raise this with the human up front rather than after a failure. On MCP clients with form elicitation, this call first confirms that setting, then prompts for phone/region and PIN and completes login by itself; if the human says the setting is off, it returns the enabling steps without starting a login (relay them verbatim, then call login again). On clients without it (e.g. Claude Desktop), phone/region come from the arguments or a persisted login, and this returns as soon as LINE issues the PIN (or reports none needed) — then call login_complete IMMEDIATELY (do not wait for the human). LINE gives ~3 minutes from PIN display to confirm on the phone; login_complete blocks past that, so calling it late only wastes that window.

Finish a passwordless login that login started on a client without form elicitation (not needed on form-elicitation-capable clients). No arguments. Call immediately after login returns — do not wait for the human. Blocks while they enter the PIN (skipped if a stored certificate is valid) and approve the device, then returns the profile. LINE's real deadline is ~3 minutes from PIN display. Errors if no login is pending.

List LINE conversations (chats, groups, rooms) with unread counts, a preview of the last message, and a human-readable name (group title, or the other party's display name for a 1:1).

Fetch messages from one LINE conversation. Text is E2EE-decrypted when keys are available; each message has fromName (resolved sender), a mediaType flag (image/video/audio/file) plus messageId for get_message_media, and mentions (LINE's raw contentMetadata.MENTION, or null — a literal "@name" in text is not itself a mention). Without before, returns the most recent count. With before (id/deliveredTime of the oldest message already seen), returns one older page — repeat to page further back.

Download and decrypt one LINE image message. Legacy alias of get_message_media restricted to images; prefer get_message_media for video/audio/file.

Download and decrypt one LINE media message of any downloadable type (image, video, audio, file). Returns image/audio MCP content, or an embedded resource blob (with filename when known) for video/file. Non-media messages (text, sticker refs, unsupported types) return an honest error naming the content type — never fabricated bytes.

One-shot unread digest: every LINE conversation with unread messages (all pages of the inbox, newest activity first), each with its most recent messages (default 10), E2EE-decrypted, with resolved sender names. Saves calling list_conversations then get_chat_messages per chat. Read-only: never marks anything read, never touches the search index; denylist-excluded conversations are omitted. Returns an empty list when nothing is unread — never fabricated.

A compact "what needs my attention" context network over the local index — you make the final call, this assembles the evidence cheaply. Nodes: connectors (people across ≥2 of your chats, with structural bridges) and relationships (per-conversation engagement, reply rhythm, recency). open: conversations whose latest message is NOT yours, ranked by how overdue they are relative to your usual reply rhythm there, each with fromName (last speaker), a preview of the latest message, overdueRatio/typicality, and a lastMessageId pointer. It carries NO message threads and makes NO judgement about addressee, nicknames, or open-request vs closing-ack — those are language understanding you do by reading each preview (a group message may be addressed to someone else, who then owns it), fetching the full thread with get_chat_messages only for the few worth it. Reads across all conversations (denylist-excluded dropped). Empty only when the index is empty.

Sends a text message to a LINE conversation immediately (not a draft). Always E2EE (pairwise for 1:1, group key for group/room); fails honestly rather than sending plaintext if the key cannot be resolved. One send per call. To @mention someone, put the visible "@name " into text AND pass a matching mentions entry — without mentions, "@name" is plain text and notifies no one. Resolve MIDs via get_group_members or find_contact first.

Sends an E2EE image to a LINE conversation immediately (encrypt → upload to OBS → send). Works for 1:1/group/room; fails honestly if the key cannot be resolved or the upload is rejected. One send per call. Provide exactly one of imagePath or imageBase64.

Sends an E2EE file attachment (any type — .docx, .pdf, .zip, …) to a LINE conversation immediately (same pipeline as send_image; the original filename is sealed end-to-end). Works for 1:1/group/room; fails honestly if the key cannot be resolved or the upload is rejected. One send per call. Provide exactly one of filePath or fileBase64; fileName is required with fileBase64 (and overrides the basename when given with filePath).

Sends an E2EE audio message to a LINE conversation immediately (same pipeline as send_file). Provide exactly one of filePath or audioBase64; optional durationMs sets the recipient player length. Works for 1:1/group/room. One send per call.

Sends an E2EE video to a LINE conversation immediately (same pipeline as send_file; uses LINE chunked video encryption so it plays and integrity-verifies on official clients). Provide exactly one of filePath or videoBase64; optional durationMs sets the scrubber length. Works for 1:1/group/room. One send per call.

Sends a location (map pin) to a LINE conversation immediately — latitude/longitude plus optional title (place name) and address. Works for 1:1/group/room. One send per call.

Shares a LINE contact card to a conversation immediately — the recipient sees a tappable card for contactMid (e.g. from find_contact or get_group_members). displayName is optional (resolved from the mid when omitted). Works for 1:1/group/room. One send per call.

Sends a LINE sticker to a conversation immediately, named by stickerId (STKID) + packageId (STKPKGID). Only OWNED stickers can be sent — get ids from search_stickers/list_stickers. Works for 1:1/group/room. One send per call.

List the sticker packages this LINE account OWNS — the only stickers it can send. Returns each package's packageId (STKPKGID), title, and version (STKVER). Get individual sticker ids via search_stickers, then send_sticker. Read-only.

Search the account's OWNED sticker packages by title and expand each match into its individual sticker ids (STKID), ready for send_sticker. Case-insensitive substring match on the package title. Read-only.

Show sticker preview images (MCP image content from the public sticker CDN) so you and the user can SEE them before sending. Give a packageId (from list_stickers/search_stickers) to preview its first stickers, or add a stickerId to preview just one. Each image is labeled with its stickerId + packageId for send_sticker. Read-only.

Send a LINE read receipt (mark a conversation read up to messageId, or the latest when omitted) — a real action the other party can see. Use ONLY when the user explicitly wants to mark a chat read; reading (get_chat_messages, get_unread_digest) and background capture never mark read. Fails honestly if there is nothing to mark.

Find LINE friends whose display name contains name (case-insensitive substring). Returns each match's mid for send_message. Raw friend-list lookup — no ranking, no fuzzy scoring.

List the authenticated user's full LINE friend list as-is (mid + displayName). No ranking, no interaction-frequency ordering.

List the members of one LINE group (mid + displayName each). Resolves persistent groups (c...); ad-hoc rooms (r...) without a group record return an honest error, never a fabricated empty list.

Renames a LINE group/chat immediately; the new name is visible to every member. Works on groups/rooms (chatId starting with c/r). One rename per call.

Invites members into a LINE group immediately. Invitees must accept before joining. Provide mids (from find_contact or get_group_members). One call per invite batch.

Removes (kicks) members from a LINE group immediately — they lose access at once (re-adding needs a fresh invite); visible to every member. Provide mids (from get_group_members). One call per removal batch.

Makes THIS LINE account leave a group immediately — it loses access to the group and its history. One leave per call.

Creates a new LINE group/room immediately with the given members (no message is sent). chatType 0 = group (invitees must accept before joining), 1 = room (members added directly); default 1. Provide name and mids (initial members, e.g. from find_contact). One create per call.

Adds a person to your LINE friends by their MID (e.g. from get_group_members or find_contact). One add per call.

Blocks a contact for your LINE account — they can no longer message you. Reversible with unblock_contact. One block per call.

Unblocks a previously blocked contact for your LINE account. Undoes block_contact. One unblock per call.

Accepts a group/chat invitation for your LINE account — you join the chat. Use for a group you were invited to (its chatId appears with invited status). One accept per call.

Adds a reaction to a LINE message, visible to the conversation. reactionType: 2 = 👍 LIKE, 3 = ❤️ LOVE, 4 = 😆 LAUGH, 5 = 😮 SURPRISE, 6 = 😢 SAD, 7 = 😡 ANGRY (default 2). One reaction per call.

Removes this account's reaction from a LINE message. Undoes a react_message. One cancellation per call.

Retracts (unsends) one of YOUR OWN LINE messages for everyone — deletes it from the conversation for all participants and CANNOT be undone. LINE allows unsending only your own messages. SAFETY GATE: you must pass confirm: true, or the call refuses so it can never fire by accident. One unsend per call.

Bulk-fetch recent messages from LINE conversations into Yomi's local cross-conversation search index (LINE has no native cross-chat search). Fetches up to perChat per chat (default 100) for chatIds, or all conversations when omitted, and best-effort embeds them for semantic search — re-running also repairs any messages still missing a vector. A background capture loop keeps the index current on its own, so call this only to force a reconcile or backfill specific chats. Undecryptable messages are skipped, not fabricated.

Search across your LINE messages. Hybrid ranking: FTS5 keyword search (covers every indexed message, so exact matches are never dropped) fused with semantic similarity when embeddings are available — the mode field reports which contributed (hybrid | semantic | keyword). If the index is empty and a session is live, it auto-collects all conversations first; a populated index searches locally with no network. Empty index with no session returns an honest notice, never a fabricated match list.

Add conversations to Yomi's privacy denylist. Excluded chats are (1) skipped by future collect_messages/search_messages auto-collect — never fetched or indexed — and (2) PURGED now: their already-indexed messages and embeddings are deleted from the local index in the same call. A real privacy action, not just a future filter. Local-index operation; works without a live LINE session.

Remove conversations from Yomi's denylist, re-allowing future capture. Does NOT restore data purged when the chat was excluded — capture resumes from empty history going forward. Local-index operation; works without a live LINE session.

List the conversations currently on Yomi's privacy denylist. Returns [{ chatId, name }]; name is best-effort resolved when a live LINE session exists, otherwise null — never a fabricated placeholder. Local-index operation; works without a live LINE session.

Return Yomi's data-capture privacy policy (the disclosure to show the user) plus the current list of excluded conversations. Call this to show the user, in concrete terms, what Yomi captures by default and how to exclude conversations.