add_favorite
ChatGPTAdd a product or dish to the user's favorites (exactly one of productId or dishId from search_foods results). Limit 200 favorites. Subject to the linked account's entitlement.
Healthcare
Overview
Kalori is a nutrition catalog for Japanese convenience stores and restaurant chains. It covers 380+ chains (Seven-Eleven, Lawson, FamilyMart, Saizeriya, Sukiya, McDonald's Japan and more) and returns calories and protein, fat, and carbohydrates for their items. Data is refreshed nightly from official chain sources, and every item carries a nutritionSource marker: official means the chain publishes the value, partial means published values completed with AI estimates, and estimated means every value is AI-estimated. A linked Kalori account is required. Personal-record features are available according to the linked account's existing access, and usage limits apply. The plugin does not offer purchases or subscription management. Main workflows: (1) find something to eat at a specific chain under a nutrition constraint, for example high-protein and under 500 kcal at Seven-Eleven; (2) look up several dishes in one call when planning a meal or a shopping list; (3) compose a meal from catalog items and render it as a meal-plan card whose totals are computed server-side from catalog values, so the numbers shown are not model estimates; (4) log what was eaten and see the day's remaining budget. Search terms, filters, and result counts are retained as the linked account's search history to improve catalog coverage; each call also records the IP address and user-agent in private operational logs that are deleted automatically after about seven days. The tools accept no precise location, chat transcripts, or third-party account identifiers, and personal records are only ever reached through the user's own linked account.
Tools
analyze_and_log_meal
ChatGPTAnalyze a meal photo or free text with Kalori's nutrition engine, saving it as an own food and logging it in one step. Besides kcal/PFC it estimates and saves 27 detailed nutrients (iron, fiber, vitamins, etc.) that feed get_nutrition_summary detailed; manual values in log_meals get no detailed nutrients, which is how to choose between them. For a photo: observe it yourself and write a detailed Japanese description in query: items and counts, portion estimates (like ご飯茶碗1杯, 唐揚げ5個), cooking method, and any numbers on a visible nutrition label. If it is identifiable as a chain product, search_foods → productId logging comes first (official values). catalogMatches in the response are catalog candidates (official values); switching to one is log_meals + delete_meal. To correct a meal just logged ("that was actually X"), pass its id as
replaceMealId instead of delete_meal and re-logging. Deleting and re-creating changes the id, which loses the taste rating, actual photo, and log date (used for streaks), and holding the old id leads to failed double deletes. replaceMealId keeps the log date and meal slot; do not pass mealDate / mealType with it (moving date or slot is update_meal). With replaceMealId the redo is only proposed, because it overwrites the record: nothing changes and no analysis runs until the user confirms in the card or at the returned link. Without it, the meal is logged directly and the result card lets the user undo it. Response cal/protein/fat/carbs are the logged values and equal the sum of components (each an absolute value for its amount, with the amount in the name, like 唐揚げ 5個 or コーラ 1缶（約350g）). The engine's amount misreadings show in those names; state the amount explicitly and re-analyze with replaceMealId. Subject to the linked account's entitlement.clear_budget
ChatGPTPropose clearing a date's custom budgets (one slot or the day's goal override); applies only after the user confirms. With mealSlot, that slot's custom budget is deleted; without it, all of that day's goal overrides are cleared (back to the base goal). Nothing changes until the user presses the button in the confirmation card or opens the returned link. Subject to the linked account's entitlement.
confirm_change
ChatGPTExecute a proposed change after the user pressed its confirmation card's button; called by the card, not the model. The token exists only in the proposal result's _meta, which the model never sees; never guess one. Runs at most once per proposal (a repeat returns the first result) and only within 1 hour of the proposal.
delete_exercise
ChatGPTPropose deleting one exercise record; nothing is deleted until the user confirms in the card or at the returned link. HealthKit-sourced rows cannot be deleted. Ids come from log_exercise / get_exercises results. Subject to the linked account's entitlement.
delete_meal
ChatGPTPropose deleting one logged meal; nothing is deleted until the user confirms in the card or at the returned link. Id from get_meals / log_meals results. The meal is checked now (a wrong id fails here), then the user decides. Subject to the linked account's entitlement.
get_day_notes
ChatGPTGet the user's free-text day notes (health, events) for a date or date range (max 90 days). Pass date (single day) or startDate+endDate; default today. Only days with a note appear in items. Reading them before reviewing meals or giving advice accounts for context the numbers don't show (health, plans, mood). Subject to the linked account's entitlement; web-registered accounts currently have grace access.
get_exercises
ChatGPTGet the user's exercise records and total burned kcal for a date (also to find ids for delete_exercise). Subject to the linked account's entitlement.
get_favorites
ChatGPTGet the user's favorite products, dishes, and own foods with nutrition, as material for meal suggestions. For requests like "pick tonight's dinner from my favorites". Items with type=product/dish can be logged directly by passing their productId/dishId to log_meals. Subject to the linked account's entitlement.
get_food_details
ChatGPTRe-fetch exact nutrition for one catalog item by its search_foods id (p_{productId} or g_{dishId}). When numbers need rechecking in a long conversation, use this instead of memory or guesses. nutritionSource official = the chain's published values; partial / estimated include AI estimates, so say they are estimates. Null nutrition means kalori has no data for the item: answer "no data" instead of guessing.
get_meals
ChatGPTGet the user's logged meals for a date or date range (max 14 days), or search all history by meal name with q. Pass date (single day) or startDate+endDate. detail=true adds components, shop, portion adjustments, and whether a photo exists (use before reviewing or editing). q searches the whole history by meal name instead of by date, with no range limit (for "when did I have that pork miso soup last week?" or "log my usual breakfast again"). On supporting hosts, meals render as a card where the user can add a taste rating (1–9) by tapping a row. Subject to the linked account's entitlement.
get_nutrition_summary
ChatGPTGet daily nutrition totals vs goals (intake kcal, PFC, exercise burn) for a date or range, with per-slot budgets. Pass date (single day) or startDate+endDate (max 90 days). A single day also returns slots (budget allocation per meal slot such as breakfast/lunch/dinner; source=auto = not logged yet): for "find something for dinner", pass that slot's cal/protein/fat/carbs to search_foods target* as is. detailed=true switches to totals of 27 nutrients (fiber, salt, iron, vitamins, etc.) with recommended intakes, for "am I getting enough X?". On supporting hosts a dashboard card renders, and the user can ask for "a meal that fits this slot" from an unlogged slot's chip. Subject to the linked account's entitlement.
get_profile_and_goals
ChatGPTGet the connected account's current entitlement, API quota, and any available profile, goals, and observed metabolism. Available data can include profile, goal PFC, weight goal, meal-slot allocation ratios, and observed metabolism (metabolism). When metabolism has trusted=false it is still being estimated, so do not state it as certain; null means insufficient logs. goal.source=manual is a goal the user fixed, and it does not track automatically. Do not assert that fields withheld by the entitlement are unset; say they are currently unavailable. This tool does not change the account's entitlement.
get_usual_meals
ChatGPTGet the user's usual meals (logged most often in the last 90 days) with nutrition and ids to re-log them. For "the usual" requests such as "log my usual coffee" or "my usual breakfast". Uses the same counts as the app's frequent-meals list (per meal slot, the top meals logged on 2 or more days), merged into one item per meal and shop across slots (recordedDays is summed over slots; mealType, lastDate, and the nutrition values come from the most recent log). q narrows by meal name or shop name (every word must match); mealType counts only that slot. To re-log, pass the item's productId, dishId, or foodId (at most one is returned) to log_meals with the returned quantity; an item with none of these ids can be logged manually with its name as mealName and the returned cal/protein/fat/carbs, which are already for the amount eaten. catalogId (p_{productId} / g_{dishId}) is the same id search_foods returns and get_food_details accepts. If nothing matches, search the catalog with search_foods using the same words. Subject to the linked account's entitlement.
get_weights
ChatGPTGet the user's weight records for a date or date range (max 90 days), to review weight-loss pace or inform meal advice. Pass date (single day) or startDate+endDate. On supporting hosts a line-chart card renders, and the user can log today's weight from the card's input field. Passing targetWeightKg (from get_profile_and_goals) draws a goal line on the chart. Subject to the linked account's entitlement.
list_shops
ChatGPTList the shop chains in Kalori (id, name, aliases) to get shop ids for search_foods / suggest_meal_set filters.
log_day_note
ChatGPTWrite the user's free-text note for a day (health, events, a quick word); one note per day, separate from meals. Everyday reports like "my stomach feels heavy today" or "had a business dinner" can be written as is. Writing the same day again overwrites it; an empty body deletes it. The result card lets the user undo it. Subject to the linked account's entitlement; web-registered accounts currently have grace access.
log_exercise
ChatGPTLog an exercise session with its burned kcal (past dates allowed via exerciseDate). A report like "ran 30 minutes, about 300 kcal" can be logged as is. Subject to the linked account's entitlement.
log_meals
ChatGPTLog meals to Kalori, up to 10 per call (batch preferred): catalog item, generic dish, own food, or manual values. When logging from a photo, log anything identifiable as a chain product via search_foods → productId first (official nutrition); use analyze_and_log_meal for dishes you cannot identify. Use manual values (mealName+cal) here when the user states the numbers. Four methods: (1) catalog product = productId from a search_foods type=product result (official nutrition; shop attached automatically); (2) generic dish = dishId from a search_foods type=dish result (standard nutrition + components added; for home cooking and non-chain eating out); (3) own food = foodId from search_my_foods (re-logs the user's own food); (4) manual = mealName + cal. Give exactly one method per item; never mix a reference id with manual PFC or components. Manual components must each have kcal/P/F/C. All manual numbers are absolute amounts for what was actually eaten, with no multiplier: for "サラダチキン40g", put the amount in the name and pass kcal/P/F/C for 40 g (components have no quantity or unit). mealDate accepts past dates (moving records) and future dates (meal planning): for "plan tomorrow's dinner and add it", batch-log with a future mealDate and it appears in the app's plan grid. A meal totaling over 10,000 kcal is rejected. The result card lets the user undo the log. Subject to the linked account's entitlement.
log_weight
ChatGPTLog the user's body weight for a date (one entry per day; logging the same day again overwrites it). Everyday reports like "65.2 kg this morning" can be logged as is. The result card lets the user undo it (restores the previous value). Subject to the linked account's entitlement.
rate_meal
ChatGPTSave the user's personal taste rating (1–9) for a logged meal; rating again overwrites it. Scale: 1 = meh … 9 = best. Omitting note keeps the previous note. Lets "that ramen earlier was a 7" be saved as is, and feeds "ベスト食事" in the app's "気づき" tab. Subject to the linked account's entitlement.
remove_favorite
ChatGPTRemove a product or dish from the user's favorites.
search_foods
ChatGPTSearch the Kalori food catalog (Japanese chain products and generic dishes) by name, shop, or kcal/PFC filters. Covers products from hundreds of Japanese chains (convenience stores and restaurants) plus thousands of generic dishes such as ramen and oyakodon; same engine as the iOS app (kana/romaji/synonyms, popularity order, semantic fallback). Pass productId (type=product) or dishId (type=dish) from results to log_meals to log with nutrition data. target* sorts by fit to the remaining budget (pass get_nutrition_summary's remaining as is). imageUrl is a public CDN product image (no auth); include it as a link when suggesting meals so the user can see the item. To look up several items (meal plan, shopping list), pass them together in
queries instead of calling repeatedly; it costs one call's wait and quota (up to 3–10 terms per call depending on the connected account's entitlement; more returns 400, so split the calls). Never pack several dishes into one query string: it is parsed as one dish name and returns 0 hits even when each exists (measured: four dish names in one query → 0 hits; one at a time → 4/4/8/3). queries returns a result array per term plus notFound (terms with no hit). query accepts only dish, product, or shop names; generic words like "メニュー", "昼ごはん", "夕食" return 0 hits. To filter only by shop or numbers (like "under 500 kcal at Saizeriya"), omit query and pass only shop / calMax / proteinMin. If a search with numeric filters returns 0, the server retries without the term and adds queryDropped and note; tell the user so. Do not guess numbers for notFound items; suggest manual logging or analyze_and_log_meal. On supporting hosts, results render as a card carousel and the user can log from a card's "＋記録" button (logging is reflected in the conversation context).search_my_foods
ChatGPTSearch the user's own foods (analyzed from photos/text in the app or registered manually) to re-log them by foodId. For "log my usual homemade smoothie": find it here and pass foodId to log_meals to re-log with nutrition and components. Without q, lists them newest first. Subject to the linked account's entitlement.
send_feedback
ChatGPTSend the user's feedback (impressions, feature requests, bug reports) to the Kalori development team. Kalori's chat integration is in development and feedback is wanted: when the user says things like "I wish it could…", "it didn't log properly", or "this search result looks wrong", suggest sending it with this tool. Always read the text back to the user and get consent before sending (never send on your own; keep the user's words rather than over-summarizing). Replies arrive in the Kalori app under マイページ → お問い合わせ, not in this chat; tell the user so. Available regardless of the current entitlement and does not consume API quota. Up to 5 per 24 hours.
set_day_budget
ChatGPTPropose overriding the daily kcal/PFC goal for one date only; it applies only after the user confirms. The base goal stays untouched, and nothing changes until the user presses the button in the confirmation card or opens the returned link. For requests like "Saturday is a cheat day, make it 2500 kcal" or "keep today light at 1500 kcal". Axes you omit stay at the base goal; an explicit null clears that axis only. Subject to the linked account's entitlement.
set_goal
ChatGPTPropose setting, changing, or creating the user's goal; it applies only after the user confirms in the card or link. Covers lose/maintain/gain, target weight, pace, activity level, and meal-slot ratios. The input is checked now (missing fields fail here, see below), and the response is a proposal with a preview of the new targets; nothing is saved until the user presses the button in the confirmation card or opens the returned link. Only the fields you pass change and the rest keep the current goal, so requests like "make my target weight 60 kg", "slower", or "lighter breakfast, bigger dinner" can be passed directly. The quantity this goal manages is
pace (% of body weight to move per week). Target weight only sets how far to go: changing it does not change the daily calorie target, only the estimated days to reach it. For "too fast / too hard" or "I want to lose faster", change pace. Calorie and PFC target numbers themselves CANNOT be set: the server derives them from the user's observed metabolism (metabolism in get_profile_and_goals). This lets the goal track reality as logs accumulate; fixing numbers directly would stop that tracking, so it is blocked on purpose. For a one-day number like "I want 1800 kcal today", use set_day_budget. Explain the proposed change with its basis using previous and derivation in the preview (baseTdeeKcal = the expenditure used as the base; baseTdeeSource: observed = back-calculated from the user's logs, mifflin = formula estimate from height/weight/age; dailyDeficitKcal = daily deficit), like "1,900 → 1,750 kcal: a 400 kcal/day deficit from an observed metabolism of 2,150 kcal". manualPreserved=true means the goal is manually fixed in the app: the intent (target weight, ratios, etc.) would be saved but the numbers stay fixed; tell the user (only the app's 「自動計算に戻す」 unfixes it; switchedFromManual is legacy and always false). Users with no goal yet (goal=null in get_profile_and_goals) can create one here (preview created=true, previous=null). Creating requires goalType and activityLevel, plus height, weight, date of birth, and sex for the calculation (heightCm / weightKg / dateOfBirth / gender). If any are missing, a 400 lists the field names in missing: ask the user for all of them at once, tell them they are used only to calculate the calorie target, then call again with them in the same call (never fill them in by guessing). Do not ask again for items already saved. The formula is binary, so let the user choose whether to calculate as male or female. heightCm etc. can also be passed when the user wants to correct them (on confirmation, overwrites the profile and recalculates the goal). Daily weight logging is log_weight. Subject to the linked account's entitlement.set_slot_budget
ChatGPTPropose a custom kcal/PFC budget for one meal slot on a date; it applies only after the user confirms. Nothing changes until the user presses the button in the confirmation card or opens the returned link (setting a slot again overwrites it). For requests like "leave 800 kcal for dinner tomorrow, I have a dinner party" or "keep lunch to 500 kcal". It shows in the app's plan grid and immediately affects get_nutrition_summary slots (source=custom) and search targets. Subject to the linked account's entitlement.
show_meal_plan
ChatGPTRender an agreed combination of catalog items as a meal-plan card with server-computed kcal/PFC totals. Totals come from exact catalog values; use this instead of showing numbers you added up yourself. Once the combination is settled with the user, present it with this tool. Items are productId (type=product) or dishId (type=dish) from search_foods / get_favorites results. targetCal etc. add fulfilment bars against the target (pass the matching slot from get_nutrition_summary slots to show how the meal fits that slot). This tool does NOT log: on supporting hosts the user can log from the card's "この献立を記録する" button (logging is reflected in the conversation context); otherwise log with log_meals.
suggest_meal_set
ChatGPTSuggest product sets (like staple + side) at one specific shop that fit a kcal/PFC target, to answer what to buy there. shopId is REQUIRED; if unknown, call list_shops first. On supporting hosts, the candidate sets render as cards and the user can log a whole set with "このセットを記録". Subject to the linked account's entitlement.
undo_change
ChatGPTUndo a direct write (meals, meal edit, exercise, weight, or day note) after the user pressed the card's undo button. Called by the card only; the token exists only in the write result's _meta, which the model never sees. Deletes what the write added, or restores the previous meal, weight, or note (refused if it changed since). Valid for 24 hours, once.
update_meal
ChatGPTEdit a logged meal (amount, slot, date, components, or shop). It applies immediately, and the result card lets the user undo it for 24 hours. The amount is one quantity multiplier ("left half" → 0.5, "it was 2 servings" → 2, "ate 1.5×" → 1.5); attach a shop afterwards ("that was from Seven-Eleven" → shopId). Nutrition is recalculated server-side. Component values are each component's absolute amount for its portion, with the amount in the name (same as manual log_meals; no multiplier). Components from get_meals detail / analyze_and_log_meal have the same shape: change only the rows to fix and send all rows. Meal kcal = sum of components × quantity. Updates that push a meal's total above 10,000 kcal are rejected. Subject to the linked account's entitlement.
Capabilities
WritesInteractive
Example Prompts
Click any prompt to copy it.
Links
App Stats
32
Tools
3
Prompts
ChatGPT
Platforms
Category
Nutrition PlanningWorks with
ChatGPT
Data refreshed daily