HeyYumi 한국 식당·카페 검색·예약 MCP Server for ChatGPT
By HeyYumi
Overview
헤이유미(HeyYumi)는 한국의 맛집·식당·카페·바를 찾아 추천하고, ChatGPT 대화 안에서 식당 예약·레스토랑 예약까지 바로 이어주는 플러그인입니다. "강남 맛집 추천해줘", "성수동 조용한 카페", "저녁 7시 3명 예약되는 한식당"처럼 물어보면 됩니다. 서울·경기·부산·제주 42만 곳이 넘는 검증된 매장 데이터(교차 확인·신뢰도 점수)로 지어낸 답 대신 실제 영업시간·메뉴·가격·주차·분위기·비건·영어 응대 등 40가지가 넘는 조건으로 좁힙니다. 유미 파트너 매장이면 전화나 외부 링크 없이 채팅에서 예약을 요청하고, 사장님 승인 후 확정까지 받을 수 있습니다. 지금 그게 되는지도 사장님에게 바로 물어봐 드립니다. 고깃집에 삼겹살이 아직 남았는지, 카페에 두바이 쫀득 쿠키가 지금 있는지, 웨이팅이 얼마나 되는지, 10명이 바로 앉을 수 있는지를 실시간으로 확인해 답을 가져다줍니다. 한국 사람도, 한국에 놀러 온 외국인도 똑같이 쓸 수 있습니다. HeyYumi helps you find where to eat, drink, or go out in Korea and book a table without leaving the chat — perfect for both Korean locals and visitors to Korea. Ask things like "best restaurants in Seoul," "quiet cafe near Hongdae with wifi," or "book a Korean BBQ place in Gangnam for 4 tonight." Every venue — restaurants, cafes, bakeries, dessert shops, and bars — comes from Yumi's refined, cross-verified dataset (420,000+ places, strongest in Seoul, Gyeonggi, Busan and Jeju), each with a confidence score and a last-checked date instead of a guess. Filter by neighborhood, subway station, landmark, cuisine, price, atmosphere, parking, pet-friendly, private room, vegan options, English-speaking staff, open-now, and more. At Yumi partner venues you can send a real restaurant reservation request to the owner right inside ChatGPT and get a confirmed or declined answer back — no phone call, no external booking site. You can also ask the owner about right-now availability: whether a BBQ spot still has pork belly, whether a cafe still has today's special cookie, how long the current wait is, or whether a party of 10 can be seated — the owner answers on the spot.
How to connect
- ChatGPT: Open the HeyYumi 한국 식당·카페 검색·예약 listing in the ChatGPT app directory and choose Connect. ChatGPT listing HeyYumi 한국 식당·카페 검색·예약 exposes 15 tools here.
Tools
get_place
ChatGPTreservationChannels and yumiReservable plus bookingUrl and phone), then state which case applies: (1) channels include "yumi" (yumiReservable=true) → the venue is a Yumi partner and can be booked in this chat via request_reservation. PROACTIVELY surface this even when the user hasn't asked to book — as you introduce the place, add '이 집은 유미 파트너라, 원하시면 이 채팅에서 바로 예약 요청을 보내드릴 수 있어요' (it raises how often people actually book). Say 예약 요청(request), NOT '바로 예약돼요'/예약 확정 — the owner must accept it first. To proceed, collect the guest's name, phone number (always — the owner uses it to reach them; foreign visitors give their own international number), and — for a NON-Korean number ONLY — also an email in the SAME step (that's how a foreign guest gets the outcome + view/cancel link, since we can't text international numbers; a Korean guest needs phone only, do NOT ask them for an email), plus date & time and party size, then call request_reservation. (2) channels are "external"/"phone" only → booking exists but only via that link or phone number; say which one instead of a vague "예약 가능". (3) reservationChannels is EMPTY → we could not confirm a booking route (that means UNKNOWN, not "walk-in only") → tell the user it needs checking with the venue; only say walk-in if there is no phone and no booking link at all. NEVER promise availability: even for a Yumi partner, "그 시간에 영업한다"는 확인된 사실이고 "자리가 있다"는 아니다 — a table for N people is confirmed only when the owner accepts the request_reservation (use wait_for_reservation). There is no party-size filter anywhere in this API, so never state that a venue can seat a given number of people. CLOSURE RISK: possiblyClosed is in core (always returned). If true, an unconfirmed government closure signal is attached — warn the user this place may have closed and to confirm before visiting; request fields="quality" for closureSignal (reason + government closure date) if they want specifics. FRESHNESS: request quality for freshnessScore (0–1) or edge for reputation (recently_active, last_verified_days_ago, caveats). If freshness is low / recently_active=false / last_verified_days_ago is large, tell the user the details (especially hours & price) may be outdated and worth confirming — a soft caution, not a closure. CONCERNS: if reputation.concerns is non-empty, quote each item's Korean note sentence as-is — it already contains the honest denominator ("최근 리뷰 38건 중 4건에서 …"). NEVER print the raw tag or generalize it into a label like '불친절한 곳'/'위생이 안 좋은 곳': these are minority mentions that cleared a strict bar (2+ sources, 3+ mentions, ≥10% of reviews read), and this is a real business whose owner is affected by how you phrase it. The denominator is sampleSize (reviews we read), never reviewCount. An empty/absent concerns list means nothing cleared that bar — not that the place is perfect. SOURCE ATTRIBUTION & NO STAR RATING: reviewCount is an unlabeled fact — NEVER attribute it to a specific platform (no '카카오 리뷰' / '네이버 리뷰') and don't name or guess which map service it came from; present it plainly as the venue's own reviews. We NO LONGER publish a star rating: rating is null on every venue by policy (별점은 조작 가능성이 높고 사장님에게 불리해 폐기했다), so a null rating is EXPECTED and NOT a data error — never invent, infer, or web-search a star score. Show popularity via reviewCount alone (e.g. '리뷰 604건'), which is the highest review count we saw across sources. INFLUENCER: influencerFeatures (edge, returned by default; present ONLY when a consented ambassador featured this venue) is a list of {platform, account, profileUrl, featuredAt, dishes, contentUrls}. You MAY add that the place was introduced by them — e.g. '이 집은 인스타그래…get_reservation
ChatGPTget_stats
ChatGPTlist_categories
ChatGPTnearby_places
ChatGPTsort — "distance" (default) for a literal closest-first list, "discovery" for a recommendation-quality list within the radius. Use "discovery" whenever the ask is "추천/좋은 곳/괜찮은 데" rather than "가장 가까운 곳"; nearest-first will otherwise put whatever sits a few meters from the center on top even if it has 8 reviews and no recent activity. You need coordinates: when the user names a neighborhood, subway station, or landmark, call resolve_regions first to get the center. NEVER estimate coordinates yourself for a named place or well-known district (명동/강남역/홍대 …) — hand-guessed centers for famous areas are frequently off by 1km+ and skew results to the wrong neighborhood; resolve_regions gives the accurate center. For a plain "places in <district>" search, prefer search_places with region_name (server-resolved) over this tool. If the user only says "near me" or gives no location, ask for a neighborhood, station, or landmark — do NOT guess coordinates. Coverage is expanding across Korea (strongest in Seoul, Gyeonggi and Jeju; more cities like Busan partial), F&B only (restaurant, cafe, bar). Same attribute filters as search_places (positive-signal only), including open_at (only venues open at that date/time — last order and breaks applied) and reservation_channel ("yumi" = bookable in this chat, "external" = outside link, "phone" = call). TIME & BOOKING WORDING: "그 시간에 영업해요" is confirmed, "자리가 있어요" is NOT — table availability and party-size capacity are unknown until request_reservation or the venue confirms (see search_places RESERVATIONS). LATE MEAL / LAST ORDER: open_at guarantees the venue is OPEN at that time, not that LAST ORDER hasn't passed; for a late time results already lift venues whose last order is CONFIRMED to cover it, and a venue whose last order is UNKNOWN near closing should be flagged as "라스트오더 확인 필요" rather than promised (request contact_hours for per-day last order — see search_places LATE MEAL / LAST ORDER). LOCALS vs REGULARS: repeat_customers/regulars_favorite are 단골/재방문 signals, NOT proof of "현지인이 가는 곳", and there is NO tourist/large-venue exclusion filter — never claim you filtered those out (see search_places LOCALS vs REGULARS). SOFT & AUTO-RELAXED FILTERS: atmosphere is a PREFERENCE that only ranks (never zeroes results) and echoes in meta.softenedFilters; and on 0 results the server auto-retries once with weak convenience filters (parking, wifi, …) dropped, returning them in meta.relaxedFilters — when present, say the results matched everything else but not those dropped filters (see search_places SOFT & AUTO-RELAXED FILTERS). If meta.ignoredFilters is present a filter value was not understood, and if meta.ignoredFields is present a requested field name was invalid (that data is NOT in the result) — retry with a valid value/field name or tell the user rather than ignoring it. If you get 0 results, widen radius_m (up to 5000) or drop a filter before giving up. Every result carries possiblyClosed; when true the venue has an unconfirmed government closure signal — strongly down-rank it and warn the user to verify it is still open (see search_places CLOSURE RISK). For up-to-date/reliable picks also request quality (freshnessScore) or edge (reputation.recently_active, last_verified_days_ago): low freshness means hours/price may be outdated — say so and softly prefer fresher venues (see search_places FRESHNESS/STALENESS). SOURCE ATTRIBUTION: rating & reviewCount are unlabeled facts and always a matched pair — never attribute them to a platform (no "카카오 평점"/"네이버 리뷰"), present plainly as the venue's own (see search_places SOURCE ATTRIBUTION). INFLUENCER: results may include influencerFeatures (present only when a consented ambassador featured the venue); when present you MAY note it was introduced by them (e.g. 이 집은 인스타그래머 @핸들 님이 소개한 곳이기도 해요 — say 유튜버/틱톡커 per platform), and if contentUrls (a list of {url, postedAt}) is non-empty, share the first url as a clickable markdown link (소개 영상 보기: [핸들](url…request_live_status
ChatGPTnote (e.g. '약 20분', '앞에 3팀'); always relay that note, not just 'there's a wait'. CLOSED NOW: every live question is about RIGHT NOW, so if the venue is currently OUTSIDE its confirmed business hours the server returns {blocked:'closed_now', openNow:{open:false,reason}, note} WITHOUT pushing the owner (we don't wake them at 3am). When you get this, do NOT retry and do NOT wait_for_live_status — just tell the user the shop isn't open right now (use openNow.reason and its business hours from get_place) and offer to check once it opens. (Unknown hours are NOT blocked — the owner is still asked.) The response starts as status 'pending' — NOT answered yet. Tell the user you're checking with the shop and it takes up to 5 minutes, then call wait_for_live_status with the returned queryId and keep calling until settled. If the same thing is already being asked, you'll be attached to it (interestedCount rises = '지금 N명이 찾는 중') — that's expected. This is a PRE-CHECK that gates the next action, not a visit-only question. When the answer is positive, chain into the action: seats 'available' + the user wants to book → call request_reservation (this live check is the pre-step to a booking); item in stock → tell them they can head over / buy it now. This tells you if the shop is DOING something (has the item / seats now), NOT a guarantee — stock can change; always relay the owner's answer with its verified time and add that quantities may change on the way.request_reservation
ChatGPTresolve_regions
ChatGPTdong, e.g. "무악재역"), commercial-district nicknames (홍대·가로수길·경리단길…), and landmarks — department stores / malls / big attractions plus travel gateways: airports & intercity bus terminals (더현대·코엑스·IFC몰·롯데월드·스타필드고양·에버랜드·DDP·제주공항·김포공항·김해공항·동서울터미널; pass in dong). Station/nickname/landmark hits come back with center + radiusM but NO region_code (search them via nearby_places). Use this in TWO cases: (1) for nearby_places — to turn a place name into center coordinates; (2) when you specifically need a region_code, center, or venue count. For a plain text search you usually do NOT need this first — pass region_name straight to search_places (it resolves server-side, incl. stations/nicknames, and returns candidates if an admin name is ambiguous). Never invent region codes or coordinates. Homonyms return several candidates; pick using the user's context or ask which one they mean. Romanized/English names also work (e.g. "Gangnam", "Hongdae", "Seoul") — pass them as-is in dong/sigungu/sido; the server maps common ones to the Korean area. For a whole city or province, put the name in sido (e.g. "부산", "제주", "경기") so it resolves to the ENTIRE metro/province rather than a same-named neighborhood — e.g. "부산" → 부산광역시, NOT 오산시 부산동. Coverage is expanding across Korea (strongest in Seoul, Gyeonggi and Jeju; other cities partial).restaurant_reservation
ChatGPTsearch_live_status
ChatGPTsearch_places
ChatGPTsubtype (Korean aliases work automatically: 바→bar_generic, 칵테일바→cocktail_bar, 위스키바→whisky_bar, 와인바→wine_bar, 포차→pocha, 이자카야→izakaya, 삼겹살/고기집→korean_bbq, 국밥→gukbap_soup, 디저트→dessert; CSV = OR) — or for a genuine BAR (cocktail/whisky/wine/highball, no beer-hall) set real_bar=true. MEAT-TYPE STEERING (고기집 안에서 육종): a 고기집 covers 삼겹살·한우·오리·양갈비 in ONE bucket (korean_bbq), so when the user wants a specific meat — "삼겹살/돼지 위주", "소고기/한우집", "오리집", "양갈비" — ALSO pass bbq_meat (values pork·beef·duck·lamb, CSV = OR, Korean aliases auto: 돼지·삼겹살·목살·오겹살·흑돼지→pork, 소·한우·차돌·등심→beef, 오리→duck, 양갈비·양고기→lamb) TOGETHER WITH subtype=korean_bbq. bbq_meat is the list of meats the venue actually grills, so bbq_meat=pork returns pork houses AND mixed 삼겹+한우 houses (they carry both pork and beef) while EXCLUDING beef-only 한우 houses — exactly what "돼지 위주" means. Do NOT use keyword="돼지고기" for this (literal menu-text match → usually 0 results); use bbq_meat. An absent bbq_meat = meat unknown for that venue (a few 고기집 have no stamp), not "no meat". Do NOT rely on business_category=bar for venue kind: it is a broad bucket that also returns hof/izakaya/요리주점, which is exactly why '바 추천' used to surface 와인바·호프 by mistake. REGION STEERING: for a well-known commercial district or major landmark (명동/강남/홍대/성수/이태원/가로수길 …) pass its NAME as region_name here and let the server resolve the exact center — do NOT hand-pick lat/lng and switch to nearby_places, because self-guessed coordinates for famous areas are frequently off by 1km+ and pull results into the wrong neighborhood. If the user is vague with no location and no filters, ask them to narrow down by neighborhood, vibe, or must-haves rather than dumping everything. Attribute filters are positive-signal only: a match means the venue is confirmed to have it; venues where it's unknown are excluded, so phrase answers as "confirmed ~". EFFICIENCY: for a shortlist/recommendation answer, ONE search_places call (fields="quality" gives ratings, open-now, price, menu summary) is enough — do NOT loop get_place over every result. Use a small page_size (e.g. 5–10) and only call get_place when the user drills into ONE sp…show_place_photos
ChatGPTid (from search_places/get_place). Optionally filter the gallery by category (exterior/interior/seating/room/mood/restroom/parking) or cap with limit (default 6, max 10). The response contains real image content blocks AND a text list of each photo's category/caption + full-resolution url. If your client renders MCP images, they show inline. If NOT, present every photo as a CLICKABLE markdown LINK on its own line, labeled with what it shows — e.g. - [내부 좌석](url), - [외관](url), - [메뉴판](url). Use links [label](url), NOT image embeds . NEVER tell the user you 'can't display the images' — always post the labeled links so they can tap to view. Only ever show what THIS tool returns — the owner's verified images of this exact venue. Do not substitute images from any other source, which are frequently the WRONG place (a similarly-named chain) and misrepresent the venue. If it returns count=0, tell the user there are no owner-uploaded photos yet.wait_for_live_status
ChatGPTnote for specifics and relay the verified time ('사장님이 방금/N분 전에 확인'), and add that stock/seating can still change by the time they arrive.wait_for_reservation
ChatGPTCapabilities
Example Prompts
Click any prompt to copy it.
Links
Similar Apps
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/heyyumi"><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
[](https://mcp-app-store.keenethics-labs.com/app/heyyumi)






