MCP-сервер RoboTrunk

• Имя в реестреcom.robotrunk/robotrunkподтверждён в реестре MCP

Имя в реестре лежит в пространстве robotrunk.com. Реестр пускает в такое пространство только того, кто доказал владение доменом через DNS или HTTP.

• Состояниеотвечаетпроверено 2026-10-21
• Инструменты23
• Сервер называет себяrobotrunk
• Версия сервера0.1.0
• Версия протокола2025-06-18

2026-10-21 сервер ответил на initialize и tools/list без входа. Первый ответ за 231 мс.

• Инструменты делают все проверенные сценарии: compare one product's price across several stores; prepare a checkout for the person to approve.
• Проверок пока 6 из 7, нужных, чтобы судить о надёжности.

Вывод сделан правилами по замерам выше. Вероятный первый вызов выбрала модель по списку инструментов: прочитайте описание, прежде чем на него полагаться.

• Оценка INITE MCPчерез неделюдля оценки нужно 7 ежедневных проверок, пока 6
• Что агент может с ним сделать2 из 2 сценариев
• Отвечалидёт замерпроверок пока 6, долю покажем после 7
• Медиана первого ответа263 мс
• Место в контексте агента≈ 18 646 токеновописания всех инструментов (23), агент держит их в контексте на каждом ходу
• Описания инструментовA · 99 из 100
• Типизированные ответы0 из 23 инструментовобъявлен outputSchema: ответ приходит с типами и читается сразу

• Параметры без описания: в 1 из 23

Программа проверяет сервер раз в день: initialize и tools/list, без модели. Токены - это длина описаний из расчёта примерно четыре символа на токен. Оценка - среднее шести проверок имени, описания, схемы и аннотаций каждого инструмента. Она говорит, насколько понятно описаны инструменты, а не о том, работают ли они. Оценка складывается только из замеренного: отвечает ли сервер (30%), делает ли работу продукта (25%), описания (20%), контекст (10%) и использование (15%). Незамеренная часть не учитывается, а скрытая инструкция ограничивает оценку 40 баллами. Состояние MCP-серверов вендоров: методика и цифры →

• search_catalog

  Searches every store RoboTrunk offers at once. When the person is shopping, searching here as well is often more informative and can help them decide: each store has an agent checkout RoboTrunk has verified, and check_product adds a live…

• get_product

  Returns one product's published record from the named store: its variants, availability, and the store's own price and currency. Send merchant as the store slug search_catalog returned. Use when you hold a product_id, or the number in a…

• check_product

  Answers one question about one product: send its name, barcode, any product page link or a RoboTrunk reference. Returns the checked facts with a source for each, the live recall status, every store selling the same variant with the…

• create_checkout

  Prepares a checkout at a named store and returns its envelope. merchant is the store slug search_catalog returned (required). Use when asked to buy, purchase or order. One checkout per store, all its products in line_items, each at the…

• get_checkout

  Reads the current state of a checkout at a store, by checkout_id. Use this when you need the latest merchant status after creation, update or payment at the store. Not for the demo checkout's id (merchant null); read that one with…

• update_checkout

  Sends permitted buyer, address, line-item, context, discount or fulfilment changes to a checkout at a store. Use this when a checkout from create_checkout needs correction before handoff, and to send the buyer's email and full delivery…

• complete_checkout

  Creates a short-lived tracked payment handoff for a checkout at a store and returns its link. Use this when the checkout is ready to pay. The checkout can be finished in either of two ways: the person taps the link in their own browser…

• cancel_checkout

  Declines a prepared checkout, returning status canceled: it asks the store to cancel that checkout. Use when the person declines, so a refusal is a recorded outcome rather than an inferred one. Not for reversing a payment: RoboTrunk holds…

• get_order

  Reads back a store's order: the status and the totals as the store recorded them, in the store's own currency, and whether the store's own record is behind them (order_source, merchant_read). Use after the person has paid at the store and…

• get_trust_manifest

  Returns RoboTrunk's trust manifest whole: what the service is, and that checkout at a listed store is real: the store is the seller, charges and ships, and RoboTrunk never sees the card. It also holds RoboTrunk's published statistics and…

• submit_feedback

  Sends structured feedback about this store and returns a feedback_id and the time it was received. Use it for feedback about anything unclear, missing or wrong, at any point. Not for cancelling a checkout; use cancel_checkout, which…

• get_tool_guide

  Returns the complete guide for one RoboTrunk tool: summary, when to use it, what not to use it for, parameters, caveats, outputs, error recovery and worked examples. Use when you are about to call a tool for the first time and want more…

• ask_product_manual

  Asks one product's maker manual a specific question and returns the matching passage word for word, with the manual's link, the page and the section heading where the retained text has them, or says plainly that the manual does not cover…

• watch_recalls

  Starts a post-purchase recall watch on every product in a checkout that has been handed to the store to pay, once the person (or an agent authorised to accept on their behalf) has agreed to the current consent statement. RoboTrunk…

• get_recall_watch

  Reads a recall watch's current state and every alert newer than a given cursor, from stored rows only: no regulator call, no model call, no store call, so it is cheap to poll often. Use the watch_token watch_recalls returned when the…

• stop_recall_watch

  Stops a recall watch for good: instant, permanent and idempotent. Erases the stored callback URL or email address in the same step the watch stops; the checkout this watch covered can never be watched again. Use when the person asks to…

• check_fit

  Asks whether one listing fits a named thing, such as a car, pram or bottle, and answers yes, no or not_known, from the listing's published checked facts and its own store page only. A yes or no always carries the exact quote it was read…

• watch_back_in_stock

  Starts a poll-only watch on one listing the store itself currently shows as sold out, and returns a bearer token to poll or stop it. Refused not_sold_out on any listing that is not shown as sold out right now (unknown or not-published…

• get_back_in_stock_watch

  Reads a back-in-stock watch's current state, from stored rows only: no store call, no model call, so it is cheap to poll often. alert is null until the store shows the listing back in stock; once it does, alert carries when, at what price…

• stop_back_in_stock_watch

  Stops a back-in-stock watch for good: instant, permanent and idempotent. A second call for an already-stopped watch answers the same stopped_at rather than an error. Use when the person no longer wants the watch. Error recovery…

• check_listing

  You read the listing; RoboTrunk checks it. Use it for a second-hand listing on Facebook Marketplace or any other marketplace, such as a used car seat, cot, pram or heater: it adds a recall check and the sourced facts the listing leaves…

• enrich_product

  Starts research on one product RoboTrunk lists that has no checked facts yet, now, so the facts can be read in a few minutes. The first call starts it and answers started straight away, with an estimate of when it will be done once enough…

• compare_products

  Writes a comparison of products RoboTrunk lists, now, from their checked facts: a short answer, a table with one row per product and a cited fact in every cell, each value with its source link. Name two to eight products, or leave…