MCP App Store
by KeenEthics
MaYa by MyShop icon

MaYa by MyShop MCP Server for Claude and ChatGPT

By My Shop Technologies Ltd.

Overview

MaYa, your personal assistant for restaurants. Ask MaYa what is near you. It shows the restaurants around an address, or around you: open now or not, today's hours, delivery terms and the minimum order. Open one to browse its menu by section, with photos and prices, and tap a dish to see its sizes, toppings and options. Meal deals are built step by step. When you know what you want, MaYa prices the basket against the restaurant's live menu and holds the price for 20 minutes. Nothing is ordered in the chat. The summary card has one button, Order directly from the restaurant. It opens the restaurant's own order page, where you sign in with your MaYa Account (your phone number and a one-time code), confirm the address on the map, add a tip or a note, and place the order. You pay by card through the restaurant's own payment link. Good to know: - No sign-in is needed to browse. - MaYa answers in the language you write in. The cards and the order page are in English or Hebrew. - Menus are collected every night. The final price is the one on the restaurant's page. - Ordering is open at participating restaurants. At the others MaYa shows the menu and says it cannot order there yet. - For restaurants in the United States and Israel.

How to connect

  1. Claude: Open the MaYa by MyShop listing in the Claude connectors directory and choose Connect. Claude listing MaYa by MyShop exposes 10 tools here.
  2. ChatGPT: Open the MaYa by MyShop listing in the ChatGPT app directory and choose Connect. ChatGPT listing MaYa by MyShop exposes 10 tools here.

Tools

build_order

ChatGPT
Price a cart against the branch's live menu and hold it for 20 minutes. Call this once the customer has said what they want. It reads the branch's menu as it is right now — prices here are the ones the till will produce, which the browsing endpoints cannot promise — and returns the line-by-line recap, the total, and the link to finish the order on the restaurant's own page. language is the language the customer writes in: the summary card and that page speak it. A basket tapped into the menu card is priced by the card's own button, which then opens the restaurant's page by itself. If the customer asks you in words to order it instead, the basket is in the card's state with a shop_product_id on every line: price what is there rather than asking them to list it again — they already tapped it in. items is a list of objects: • shop_product_id — REQUIRED, the item_id from get_menu_items. Never invent one and never carry an id over from another business: the same number means a different dish elsewhere. • quantity — defaults to 1. • size — the size NAME as the menu spells it. Required when the dish has several priced sizes and no default; you will be told if it is. • toppings — topping NAMES to ADD, not ids. They are resolved here against that exact dish. An unrecognised name refuses the line and comes back with the names that do exist. A bare name puts the topping on the whole dish. To put it on part of one, send {"name": "Olives", "on": "half_a"} — "half_a" is the first half the customer described, "half_b" the other, and "quarter_a1" / "quarter_a2" / "quarter_b1" / "quarter_b2" are the quarters inside those halves. Only a dish that divides offers them: get_item's rules says how many slices it has, and a placement it does not offer is refused with the ones it does. "Half mushrooms, half olives" is two toppings, one on each half — not two dishes. A name listed twice is two of that topping ("double cheese"), where get_item gives it a max of two or more. A topping on part of a dish is charged for the part it covers, wherever the dish divides its topping price, and inside a deal it spends only that share of the step's free allowance. • option — the dish's own choice by name (which coffee, which dough). One name only. A dish that offers choices and is ordered without one comes back asking, with the list — nothing upstream says whether the choice is mandatory, so it is put to the customer rather than guessed. • without — ingredients to take OFF what the dish already comes with. REMOVALS ONLY: every name here is something the customer does NOT want. Listing an ingredient they asked to keep takes it off their plate. • note — a free-text kitchen note ("well done", "cut in eight"). Use without for "no onions", not this: a note is read by a human, the removal is applied by the till. • pack — for a meal deal, the steps answered. See below. order_type is "delivery" or "pickup" and must be one the branch offers; get_business lists them. Meal deals. When get_item shows build_steps, the dish is a deal and those steps are its choices. Ask the customer step by step, then send one pack entry per step: its level_id plus a picks list. • Every step marked required has to be answered. Sending the deal with none of them still buys it — at the full price, with nothing in it. • picks_needed is how many times the customer is asked, not a set size: 3 means three picks off the same list, one at a time. Send exactly that many. Repeats are fine unless the step says no_repeats. • Take each pick's item_id from THAT step's own choices. A choice that belongs to a different step of the same deal is refused here, because the branch itself would accept it and then apply the wrong step's pricing. • The deal's price already covers its choices. Where a step lists free_toppings, that many toppings are included per pick — ask for the ones the customer wants first, since the ones asked for first are the ones covered, and the allowance starts over at every pick. • Do not offer paid extras inside a deal unprompted. If the customer ask…

find_businesses

ChatGPT
Find businesses near a place, nearest first. Pass near only when the customer said where: a plain address — the catalog geocodes it, so "Rothschild 1, Tel Aviv" is enough and you should not geocode it yourself — or coordinates as "lat, lon". When they did not say, leave it out: a host that shares the customer's approximate location is searched from there, and the result then says approximate_location: true, so speak of "near you", never of a street. name finds a branch the customer named without saying where. It matches the name as written, so a shorter piece of it matches more. Such a result has NO distance and NO delivery answer: there was no address to measure from, and both of those are answers about an address — ask where they are before quoting either. Never put in a city you picked yourself. A guessed location does not fail — it searches somewhere else and answers about whatever stands there, which reads as an answer about the branch they asked for. With neither a place nor a name, and no location from the host (the result is then no_location), ASK THEM WHERE, once. categories is a list and behaves as OR: ["pizza", "cafe"] finds either. Use the tags from list_categories. radius is in unit, which is "mi" or "km". delivers_only keeps only businesses that deliver to the searched point — note that it also drops the ones whose delivery area we cannot determine, so it narrows more than it appears to. open_only keeps just the branches serving right now — use it when the question is who is open, because the list is otherwise unfiltered and capped by limit, so the open one can sit past the cut while the page fills with shut ones. Without the flag, read open_now on each result rather than leaving the customer to work it out; a branch closed now is still worth naming when they asked what exists nearby. A branch whose hours were never collected counts as open, because no hours is not evidence of being shut. Each result carries uid, which is what every other tool here takes. It is a handle for your calls, never a word for the customer: not in a reply, and not in a reply or a button you suggest they tap. Call the branch by its name and town; the uid stays yours to use. delivers_here is true, false, or "unknown" — the third is a real answer for merchants who price delivery by named area, and must not be reported as "no". offers are order-level offers; offers on individual dishes are returned by the menu tools instead. delivery is the fee for the address that was searched, not a generic figure, whenever the merchant prices delivery themselves. Some hand delivery to a courier that prices each job at checkout; for those it says so, and the figures it mentions are indicative — do not present them as the price. A missing delivery means the merchant set no price at all, which is not the same as free. delivery_time is what the merchant promises. An empty result does not prove the area is empty: businesses appear only once their merchant has published them. Try a wider radius or another category before telling the user there is nothing there. Search once per question. The result is drawn as a card of places, so a second search in the same answer draws the same places twice, and listing them in words repeats the card: reply in one short sentence. In ChatGPT each place on the card has an Order button that opens its menu; in Claude the card has none, so ask which place they would like. Offer only what the search returned and what the customer asked for: no other kind of food, and no place the search did not find. language is the language the customer writes in; the card shows its words in it.

get_business

ChatGPT
Everything the catalog holds about one business: phone, order types, minimum order, opening hours, preparation time, site, logo, loyalty status and order-level offers. uid comes from find_businesses. Call this only when the search result lacks something you need — the search already returns name, address, distance, categories, offers, the delivery fee and the delivery time. opening_hours is per order type and per weekday, as the merchant keeps it. Two warnings about reading it. A whole week of 00:00–23:59 looks identical whether the branch really is open around the clock or nobody ever filled the form in, so do not announce "open 24 hours" with confidence. And windows the merchant saved without choosing an order type arrive under withoutOrderType — they are their hours, but which order types they govern is not known, so do not attach them to delivery or collection yourself.

get_cart

ChatGPT
What the customer built in the menu card and has not ordered yet. Call this the moment they refer to a basket you did not assemble — "do you see what I picked", "is that everything", "how much is it". The card cannot hand you its contents: it pushes them and the hosts do not deliver that to a model, verified against a live order where the card showed a total and nothing reached the assistant. This reads the same cart back out of the catalog instead. A cart exists here only after the customer presses the order button in the card, which is what prices it, and only for this conversation; before that there is nothing to see, and saying so is better than asking them to type their order out again. It holds for twenty minutes and carries the same checkout_url as the priced basket, so nothing has to be rebuilt. language is the language the customer writes in.

get_item

ChatGPT
Everything choosable on one dish, with prices: sizes, included toppings, extra toppings, option groups, and — for meal deals — each build step with its choices. item_id comes from get_menu_items and is scoped to that business; the same number means a different dish elsewhere. Like the uid, it is never a word for the customer. Use this to price an order accurately or to answer "what can I put on it". Toppings priced "free" are genuinely included at no charge. merchant_notes is the merchant's own wording and sometimes states a constraint found nowhere else. build_steps appears for meal deals — anything the menu marked is_meal_deal has to come through here before it can be ordered, since its steps and their level_ids exist nowhere else. Each step carries picks_needed (how many times the customer is asked, one pick at a time), whether it is required, how many toppings it includes, and its choices with the item_id each pick needs. A choice may carry options of its own — ask for it as a dish in its own right if you need them. This is the most expensive call here: it reads the stored menu document. Call it for the dishes that matter, not for a whole section. A customer who wants to build or change a dish from the menu card does it in the card: tapping Add on the dish walks them through these same choices. Say so rather than reading the options out. image_url is the merchant's own photograph of the dish. Do not describe a picture: say what the menu says. Merchants fill these slots with whatever they have — a topping called Mushrooms can arrive as the restaurant's logo.

get_item_config

ChatGPT
The structure behind one dish, for the menu page to draw controls from. App-only. get_item is the model's view of the same dish; this one keeps the branch's structure — sizes, option list, topping limits, how many slices the dish is cut into, and a pack's levels with what may be chosen at each — because a configurator has to enforce what a reader only has to understand. within reads a choice that belongs to a deal: those have no menu entry of their own, so the deal that offers the choice has to be named. language is the language of the card asking.

get_menu

ChatGPT
Return one business's menu as a list of SECTIONS — not dishes. Takes the branch's uid, or its name when that is all the customer gave. Each section carries its name, how many dishes it holds and its price range. To see dishes, call get_menu_items with a section from here. Its identifier is a handle for that call, never a word for the customer. When the customer came for a kind of dish ("pizza", "sushi"), call get_menu_items once, for the section that holds it, and stop there: they asked for pizza, not for the menu. Do not fetch the other sections, and do not list the sections or their dishes in your reply. This is deliberate: a menu is 75 dishes on average and runs to thousands at the tail, and pulling all of them to answer "do they have salads" spends the context you need for the answer. An empty sections list means no menu has been collected for this business yet — it does not mean the restaurant has nothing to eat, and should be reported as "we do not have their menu", not as "they have no menu". collected_at is when the menu was last read from the merchant. Mention it when quoting prices.

get_menu_items

ChatGPT
List dishes in one section of a menu, or dishes whose name matches a query. One of section or query is required. section is the identifier from get_menu. query is a substring match on the dish name — useful for "do they have gluten free pizza" without walking every section. Each dish carries item_id for get_item (a handle for your calls, never a word for the customer), its price, a photo, and any offers on that specific dish (offers) such as being on sale. choices counts the toppings and options available — when it is present, get_item will return the full list with prices. unavailable: true means the dish is on the menu but out of stock or seasonally off; it is still worth mentioning as unavailable rather than hiding. open_now and today_hours are the branch's state today, and order_offers its order-level offers. language is the language the customer writes in; the card shows its words in it. can_order false means this restaurant does not take orders through MaYa: the card has no order button, so say that, and do not price an order for it. The result is drawn as a card the customer orders from: the dishes with their photos and prices, each dish's choices (sizes, toppings, halves and quarters, a deal's steps), the basket, and the button that orders directly from the restaurant. So your reply is one short sentence that points to the card. Do not list its dishes again, do not walk through a dish's choices in words, and do not suggest a reply that does either. One call per answer: a second call in the same turn draws a second card or replaces the first.

list_categories

ChatGPT
List the business categories the catalog can filter by, with how many businesses each has and the other words that map to it. Worth calling before a first search when the user's word is not obviously one of ours: "coffee", "פיצה" and "pizzeria" all resolve, but an invented category matches nothing and returns an empty search that looks like "there are no cafés nearby".

price_cart

ChatGPT
build_order, taking the basket in the shape the card already holds. App-only, and the pair of the cart the draft comes back with: the page sends its lines and gets its lines. The transcription between a basket line and an order item lives on this side because both halves of it do, and a copy of it shipped inside the page is a copy that drifts. language is the language of the card asking; the summary card and the checkout link follow it.

build_order

Claude
Price a cart against the branch's live menu and hold it for 20 minutes. Call this once the customer has said what they want. It reads the branch's menu as it is right now — prices here are the ones the till will produce, which the browsing endpoints cannot promise — and returns the line-by-line recap, the total, and the link to finish the order on the restaurant's own page. language is the language the customer writes in: the summary card and that page speak it. A basket tapped into the menu card is priced by the card's own button, which then opens the restaurant's page by itself. If the customer asks you in words to order it instead, the basket is in the card's state with a shop_product_id on every line: price what is there rather than asking them to list it again — they already tapped it in. items is a list of objects: • shop_product_id — REQUIRED, the item_id from get_menu_items. Never invent one and never carry an id over from another business: the same number means a different dish elsewhere. • quantity — defaults to 1. • size — the size NAME as the menu spells it. Required when the dish has several priced sizes and no default; you will be told if it is. • toppings — topping NAMES to ADD, not ids. They are resolved here against that exact dish. An unrecognised name refuses the line and comes back with the names that do exist. A bare name puts the topping on the whole dish. To put it on part of one, send {"name": "Olives", "on": "half_a"} — "half_a" is the first half the customer described, "half_b" the other, and "quarter_a1" / "quarter_a2" / "quarter_b1" / "quarter_b2" are the quarters inside those halves. Only a dish that divides offers them: get_item's rules says how many slices it has, and a placement it does not offer is refused with the ones it does. "Half mushrooms, half olives" is two toppings, one on each half — not two dishes. A name listed twice is two of that topping ("double cheese"), where get_item gives it a max of two or more. A topping on part of a dish is charged for the part it covers, wherever the dish divides its topping price, and inside a deal it spends only that share of the step's free allowance. • option — the dish's own choice by name (which coffee, which dough). One name only. A dish that offers choices and is ordered without one comes back asking, with the list — nothing upstream says whether the choice is mandatory, so it is put to the customer rather than guessed. • without — ingredients to take OFF what the dish already comes with. REMOVALS ONLY: every name here is something the customer does NOT want. Listing an ingredient they asked to keep takes it off their plate. • note — a free-text kitchen note ("well done", "cut in eight"). Use without for "no onions", not this: a note is read by a human, the removal is applied by the till. • pack — for a meal deal, the steps answered. See below. order_type is "delivery" or "pickup" and must be one the branch offers; get_business lists them. Meal deals. When get_item shows build_steps, the dish is a deal and those steps are its choices. Ask the customer step by step, then send one pack entry per step: its level_id plus a picks list. • Every step marked required has to be answered. Sending the deal with none of them still buys it — at the full price, with nothing in it. • picks_needed is how many times the customer is asked, not a set size: 3 means three picks off the same list, one at a time. Send exactly that many. Repeats are fine unless the step says no_repeats. • Take each pick's item_id from THAT step's own choices. A choice that belongs to a different step of the same deal is refused here, because the branch itself would accept it and then apply the wrong step's pricing. • The deal's price already covers its choices. Where a step lists free_toppings, that many toppings are included per pick — ask for the ones the customer wants first, since the ones asked for first are the ones covered, and the allowance starts over at every pick. • Do not offer paid extras inside a deal unprompted. If the customer ask…

find_businesses

Claude
Find businesses near a place, nearest first. Pass near only when the customer said where: a plain address — the catalog geocodes it, so "Rothschild 1, Tel Aviv" is enough and you should not geocode it yourself — or coordinates as "lat, lon". When they did not say, leave it out: a host that shares the customer's approximate location is searched from there, and the result then says approximate_location: true, so speak of "near you", never of a street. name finds a branch the customer named without saying where. It matches the name as written, so a shorter piece of it matches more. Such a result has NO distance and NO delivery answer: there was no address to measure from, and both of those are answers about an address — ask where they are before quoting either. Never put in a city you picked yourself. A guessed location does not fail — it searches somewhere else and answers about whatever stands there, which reads as an answer about the branch they asked for. With neither a place nor a name, and no location from the host (the result is then no_location), ASK THEM WHERE, once. categories is a list and behaves as OR: ["pizza", "cafe"] finds either. Use the tags from list_categories. radius is in unit, which is "mi" or "km". delivers_only keeps only businesses that deliver to the searched point — note that it also drops the ones whose delivery area we cannot determine, so it narrows more than it appears to. open_only keeps just the branches serving right now — use it when the question is who is open, because the list is otherwise unfiltered and capped by limit, so the open one can sit past the cut while the page fills with shut ones. Without the flag, read open_now on each result rather than leaving the customer to work it out; a branch closed now is still worth naming when they asked what exists nearby. A branch whose hours were never collected counts as open, because no hours is not evidence of being shut. Each result carries uid, which is what every other tool here takes. It is a handle for your calls, never a word for the customer: not in a reply, and not in a reply or a button you suggest they tap. Call the branch by its name and town; the uid stays yours to use. delivers_here is true, false, or "unknown" — the third is a real answer for merchants who price delivery by named area, and must not be reported as "no". offers are order-level offers; offers on individual dishes are returned by the menu tools instead. delivery is the fee for the address that was searched, not a generic figure, whenever the merchant prices delivery themselves. Some hand delivery to a courier that prices each job at checkout; for those it says so, and the figures it mentions are indicative — do not present them as the price. A missing delivery means the merchant set no price at all, which is not the same as free. delivery_time is what the merchant promises. An empty result does not prove the area is empty: businesses appear only once their merchant has published them. Try a wider radius or another category before telling the user there is nothing there. Search once per question. The result is drawn as a card of places, so a second search in the same answer draws the same places twice, and listing them in words repeats the card: reply in one short sentence. In ChatGPT each place on the card has an Order button that opens its menu; in Claude the card has none, so ask which place they would like. Offer only what the search returned and what the customer asked for: no other kind of food, and no place the search did not find. language is the language the customer writes in; the card shows its words in it.

get_business

Claude
Everything the catalog holds about one business: phone, order types, minimum order, opening hours, preparation time, site, logo, loyalty status and order-level offers. uid comes from find_businesses. Call this only when the search result lacks something you need — the search already returns name, address, distance, categories, offers, the delivery fee and the delivery time. opening_hours is per order type and per weekday, as the merchant keeps it. Two warnings about reading it. A whole week of 00:00–23:59 looks identical whether the branch really is open around the clock or nobody ever filled the form in, so do not announce "open 24 hours" with confidence. And windows the merchant saved without choosing an order type arrive under withoutOrderType — they are their hours, but which order types they govern is not known, so do not attach them to delivery or collection yourself.

get_cart

Claude
What the customer built in the menu card and has not ordered yet. Call this the moment they refer to a basket you did not assemble — "do you see what I picked", "is that everything", "how much is it". The card cannot hand you its contents: it pushes them and the hosts do not deliver that to a model, verified against a live order where the card showed a total and nothing reached the assistant. This reads the same cart back out of the catalog instead. A cart exists here only after the customer presses the order button in the card, which is what prices it, and only for this conversation; before that there is nothing to see, and saying so is better than asking them to type their order out again. It holds for twenty minutes and carries the same checkout_url as the priced basket, so nothing has to be rebuilt. language is the language the customer writes in.

get_item

Claude
Everything choosable on one dish, with prices: sizes, included toppings, extra toppings, option groups, and — for meal deals — each build step with its choices. item_id comes from get_menu_items and is scoped to that business; the same number means a different dish elsewhere. Like the uid, it is never a word for the customer. Use this to price an order accurately or to answer "what can I put on it". Toppings priced "free" are genuinely included at no charge. merchant_notes is the merchant's own wording and sometimes states a constraint found nowhere else. build_steps appears for meal deals — anything the menu marked is_meal_deal has to come through here before it can be ordered, since its steps and their level_ids exist nowhere else. Each step carries picks_needed (how many times the customer is asked, one pick at a time), whether it is required, how many toppings it includes, and its choices with the item_id each pick needs. A choice may carry options of its own — ask for it as a dish in its own right if you need them. This is the most expensive call here: it reads the stored menu document. Call it for the dishes that matter, not for a whole section. A customer who wants to build or change a dish from the menu card does it in the card: tapping Add on the dish walks them through these same choices. Say so rather than reading the options out. image_url is the merchant's own photograph of the dish. Do not describe a picture: say what the menu says. Merchants fill these slots with whatever they have — a topping called Mushrooms can arrive as the restaurant's logo.

get_item_config

Claude
The structure behind one dish, for the menu page to draw controls from. App-only. get_item is the model's view of the same dish; this one keeps the branch's structure — sizes, option list, topping limits, how many slices the dish is cut into, and a pack's levels with what may be chosen at each — because a configurator has to enforce what a reader only has to understand. within reads a choice that belongs to a deal: those have no menu entry of their own, so the deal that offers the choice has to be named. language is the language of the card asking.

get_menu

Claude
Return one business's menu as a list of SECTIONS — not dishes. Takes the branch's uid, or its name when that is all the customer gave. Each section carries its name, how many dishes it holds and its price range. To see dishes, call get_menu_items with a section from here. Its identifier is a handle for that call, never a word for the customer. When the customer came for a kind of dish ("pizza", "sushi"), call get_menu_items once, for the section that holds it, and stop there: they asked for pizza, not for the menu. Do not fetch the other sections, and do not list the sections or their dishes in your reply. This is deliberate: a menu is 75 dishes on average and runs to thousands at the tail, and pulling all of them to answer "do they have salads" spends the context you need for the answer. An empty sections list means no menu has been collected for this business yet — it does not mean the restaurant has nothing to eat, and should be reported as "we do not have their menu", not as "they have no menu". collected_at is when the menu was last read from the merchant. Mention it when quoting prices.

get_menu_items

Claude
List dishes in one section of a menu, or dishes whose name matches a query. One of section or query is required. section is the identifier from get_menu. query is a substring match on the dish name — useful for "do they have gluten free pizza" without walking every section. Each dish carries item_id for get_item (a handle for your calls, never a word for the customer), its price, a photo, and any offers on that specific dish (offers) such as being on sale. choices counts the toppings and options available — when it is present, get_item will return the full list with prices. unavailable: true means the dish is on the menu but out of stock or seasonally off; it is still worth mentioning as unavailable rather than hiding. open_now and today_hours are the branch's state today, and order_offers its order-level offers. language is the language the customer writes in; the card shows its words in it. can_order false means this restaurant does not take orders through MaYa: the card has no order button, so say that, and do not price an order for it. The result is drawn as a card the customer orders from: the dishes with their photos and prices, each dish's choices (sizes, toppings, halves and quarters, a deal's steps), the basket, and the button that orders directly from the restaurant. So your reply is one short sentence that points to the card. Do not list its dishes again, do not walk through a dish's choices in words, and do not suggest a reply that does either. One call per answer: a second call in the same turn draws a second card or replaces the first.

list_categories

Claude
List the business categories the catalog can filter by, with how many businesses each has and the other words that map to it. Worth calling before a first search when the user's word is not obviously one of ours: "coffee", "פיצה" and "pizzeria" all resolve, but an invented category matches nothing and returns an empty search that looks like "there are no cafés nearby".

price_cart

Claude
build_order, taking the basket in the shape the card already holds. App-only, and the pair of the cart the draft comes back with: the page sends its lines and gets its lines. The transcription between a basket line and an order item lives on this side because both halves of it do, and a copy of it shipped inside the page is a copy that drifts. language is the language of the card asking; the summary card and the checkout link follow 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/maya-by-myshop"><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/maya-by-myshop)