MCP App Store
by KeenEthics
Pet Hardware icon

Pet Hardware MCP Server for ChatGPT

By Pet Hardware

Overview

Pet Hardware is a Czech online shop with hardware for making collars, leashes and harnesses: snap hooks, buckles, D-rings, rivets, webbing and leathercraft tools. With this app ChatGPT can search the catalogue by type, material, size and colour, show live stock levels and quantity-based prices, list shipping methods and delivery estimates for a destination country, suggest matching parts for a project and answer questions from the published FAQ in seven languages. The app only reads catalogue data; it never places orders or asks for personal data. Purchases are completed on pethardware.com.

How to connect

  1. ChatGPT: Open the Pet Hardware listing in the ChatGPT app directory and choose Connect. ChatGPT listing Pet Hardware exposes 17 tools here.

Tools

check_product_quantity_availability

ChatGPT
Check whether a requested quantity can be fulfilled now without inventing unavailable stock data. availability.fulfillment_status: "fully_available" (can ship the full quantity now), "partially_available" (some stock, less than requested — see missing_qty), "unavailable" (stock_qty is 0), or "unknown" (exact quantity not tracked for this item — say so instead of guessing a number; do not treat "unknown" as either available or not). Parameters: mpn — required product MPN/SKU quantity — required, requested quantity (must be >= 1) lang — output language warehouse — reserved for future multi-warehouse support, currently informational only Response includes a ready-to-use "assistant_answer" sentence when stock_qty is known. Use get_product instead if you only need the raw availability status/stock without the quantity-fulfillment interpretation.

compare_products

ChatGPT
Compare two or more products by MPN/SKU or ID and return differing fields and parameters. differences[] lists each field whose values differ; for a numeric field "highest_value" lists the product(s) with the largest raw number (several when tied) — a fact, not a recommendation, so never call that product "the best". fields: any of id, mpn, sku, gtin, title, category, url, image_url, availability, price, price_tiers, material, finish, load_capacity_kg, params, brand, is_variant, data_quality, customs (HS/TARIC codes). Default (unset): price, availability, load_capacity_kg, params, url.

get_category_aggregates

ChatGPT
Return short aggregates over a category or filtered product set: counts, availability, groups and stats. Use this instead of get_category_products when the customer's question is about "how many" or "what's the price range", not about listing individual products — it's cheaper and the response is not paginated. Parameters: slug — category slug filters / user_filters / numeric_filters — same shapes as get_category_products (system-name-keyed, human-phrased, and numeric-range filters respectively) group_by — parameter system names (or "color"/"material"/"finish") to break the aggregate down by, e.g. ["COLOR", "WIDTH"]. Use get_param_values instead for the values of a single parameter without the full aggregate payload. include_price_stats — include min/max/avg price, default True include_load_capacity_stats — include load-capacity stats where applicable, default True in_stock_only — restrict aggregates to in-stock products match_mode / auto_correct_filters — same as get_category_products A user_filter alias covers several fields ("strap_width" is the strap WIDTH, the INNER_WIDTH of a ring and the EYE_DIAMETER of a snap hook), and each product is judged by the first of them it actually has. Fields where the value does not exist at all are listed in "excluded_by_filter" (with "nearest_values" where numeric) together with the products they excluded — say so instead of claiming those products are out of stock. A non-empty "excluded_by_filter" does NOT mean the query was only partly satisfied: with a multi-field alias one field is used and the others routinely end up here. Judge whether the query was satisfied by "exact_matches"/"fallback_used", not by this list.

get_category_filter_schema

ChatGPT
Return available filters and aliases for a category without returning products. By default values are omitted (some parameters like COLOR have 100+ distinct values) — set include_values=True to embed them inline, or call get_param_values for one specific parameter's values instead.

get_category_photo_coverage

ChatGPT
List the products in a category that still need fashion photos, fewest first. This is the tool for "which products have the fewest photos" — do NOT try to answer that by calling get_product for each product and counting "gallery_photos". That field is capped at five and mixes real-life photos with technical close-ups, so it cannot rank photo coverage: a product showing four technical shots and no fashion photo looks covered but is exactly the one that needs shooting. Counting is per VARIANT GROUP, not per single product: a photo attached to one size shows on the whole product page (in one real category 81 % of photo links sit on a variant, not on the main product), and a photo shared by two variants is counted once. "fashion_photos" are public photos marked as belonging to the Inspirium gallery — the real-life shots. "technical_photos" are the rest, returned as context only; they never count towards the threshold. Products that already have "enough_photos" (default 5) are left out entirely and only counted in "covered_count" — above that threshold there is nothing left to shoot, so the exact number is not computed. Parameters: slug — category slug; subcategories are included, because a category is often just a folder whose products hang on its children enough_photos — coverage threshold, default 5 limit/offset — pagination over the products that need photos, default 50

get_category_products

ChatGPT
Return products of a category with normalized filters and pagination. category_slug = alias of slug. Unknown slug: search_products. Valid filters/aliases: get_category_filter_schema. Counts only: get_category_aggregates. Filters (all optional): - param_filters {"COLOR": "cerna", "WIDTH": "20MM"}; user_filters {"color": "black", "strap_width": "20 mm"}. A value may be a LIST (["black", "silver"]) or a RANGE ({"min": "16 mm", "max": "25 mm"} or "16-25 mm"). - numeric_filters {"load_capacity_kg": {"gte": 250}, "price.czk.amount": {"lte": 100}} - range_mode: "snap" (default) widens bounds to available values ("17-24 mm" over 16/20/25 mm gives 16–25 mm), "strict" keeps only values inside. If applied_filters.<param>.range_was_expanded is true, quote applied_range, not the requested_range, and tell the customer. - match_mode: exact / normalized (default) / fuzzy. Fuzzy corrects a value only at high similarity, never between near-tied candidates; otherwise FILTER_VALUE_NOT_FOUND lists plausible "suggestions" (empty: call get_param_values). Numeric values are never auto-corrected: offer the error's "nearest_values". - include: category, products (default), aggregates; limit (default 20), offset. - fields: id, mpn, sku, gtin, title, category, url, image_url, availability, price, price_tiers, material, finish, load_capacity_kg, params; brand, is_variant, data_quality, customs (HS/TARIC) only if requested. "filter_url": shareable link with the filters pre-selected — send it. A user_filter alias covers several fields ("strap_width" = strap WIDTH, ring INNER_WIDTH, snap hook EYE_DIAMETER); each product is judged by the first field it has. Fields where the value does not exist go to "excluded_by_filter" with the products they excluded — say so, don't call them out of stock. A non-empty list is NOT a partial match; judge by "exact_matches"/"fallback_used". Such an alias is left out of "filter_url", so the link is WIDER than the result ("filter_url_missing_filters") — never present it as exact.

get_compatible_products

ChatGPT
Find products that go together with the given product across categories — the parts a customer needs after finding one component (e.g. a 6 mm rivet also needs a 6 mm rivet setting tool and matching dies; a 20 mm webbing needs a 20 mm buckle and D-ring). Each item carries "basis" and "confidence" — the groups are NOT equivalent. "catalog_explicit" (1.0) means the catalogue itself lists the pair as compatible, "curated" (0.9) that a human picked it, "size_and_finish" (0.8) that they share the functional size and finish, "order_history" (0.5) only that customers often buy them together — correlation, not a technical guarantee. Never present an order_history match as "these fit together". Returns four labelled groups, not one mixed list: - works_with: explicit compatibility from the catalog's own "Use for" data (works both ways — a rivet setting tool lists its rivets, and a rivet resolves back to the tools that list it). The strongest, most reliable signal when present. - same_functional_size: other categories' decorative-hardware products sharing the same functional size (inner width/diameter), finish, material — e.g. ring + buckle + loop for the same webbing. Only applies to finished hardware, not tools (a tool's numeric size doesn't imply compatibility with unrelated same-sized products). - frequently_bought_together: products real customers actually bought in the same order. - manually_curated: editor-picked companion products for this specific MPN. fields: any of id, mpn, sku, gtin, title, category, url, image_url, availability, price, price_tiers, material, finish, load_capacity_kg, params, brand, is_variant, data_quality, customs (HS/TARIC codes). Default (unset) returns a standard subset without brand/is_variant/data_quality/customs — use this to trim the response when you don't need full pricing/customs detail for every compatible product. Use this after search_products/get_product once you have a specific MPN, to assemble a full kit instead of guessing compatible parts from dimensions alone.

get_customer_answer

ChatGPT
Return one grounded support answer for a customer's question, or null when no FAQ entry answers it. The answer TEXT is extracted verbatim from a FAQ page, never generated, so it cannot add facts that are not on that page. An exact FAQ-title match is lexical; anything else, including a paraphrase, goes to a small LLM that picks an entry id (or "none") from every FAQ entry in the shop, seeing each title with the opening of its answer — it writes no text of its own (see search_faq for the underlying source). Parameters: question — required, the customer's question in natural language lang — output language: cs / en / de / fr / pl / es / ru context_tags — optional keywords to disambiguate (a product category or order type the customer already mentioned) include_sources — include the matched FAQ links, default True answer_style — concise (~700 chars, cut on a word boundary) / full, default concise max_sources — how many FAQ candidates to consider, default 5 Response semantics: confidence — "high" (worded exactly like the FAQ title), "medium" (chosen by the LLM selector by meaning; the common case), "low" only together with answer=null. answer=null — no FAQ entry answers it (or the selector is temporarily unavailable, see data_quality.warnings). Say you don't have this in the FAQ instead of guessing; always check for null first. sources[].score — null for a selector pick; with "medium" sources hold only that entry. Latency: "medium" adds one internal LLM call (a few seconds) — do not retry early. Use search_faq to inspect or rank several FAQ entries yourself.

get_param_values

ChatGPT
Return distinct values (optionally with counts) for one category parameter, such as which COLOR values actually occur once other filters are applied. Parameters: slug — category slug param — the parameter's system name (e.g. COLOR, WIDTH), default COLOR. Use get_category_filter_schema first to see available parameter names for the category if you don't already know them. filters — param-name-keyed filters, e.g. {"WIDTH": "20MM"} user_filters — human-phrased equivalent, e.g. {"strap_width": "20 mm"} in_stock_only — restrict to in-stock products before counting values include_counts — include a product count per value, default True match_mode — same as get_category_products (fuzzy thresholds included) auto_correct_filters — snap near-miss filter values to the closest real one A FILTER_VALUE_NOT_FOUND error for a numeric parameter carries "nearest_values": the closest values that really exist (e.g. asking for 21 mm returns 20 mm and 25 mm). Offer those to the customer — numeric filters are never auto-corrected to a guess. A user_filter alias covers several fields ("strap_width" is the strap WIDTH, the INNER_WIDTH of a ring and the EYE_DIAMETER of a snap hook), and each product is judged by the first of them it actually has. Fields where the value does not exist at all are listed in "excluded_by_filter" (with "nearest_values" where numeric) together with the products they excluded — say so instead of claiming those products are out of stock. A non-empty "excluded_by_filter" does NOT mean the query was only partly satisfied: with a multi-field alias one field is used and the others routinely end up here. Judge whether the query was satisfied by "exact_matches"/"fallback_used", not by this list. Use get_category_filter_schema instead if you need the schema of ALL filterable parameters at once rather than the values of one specific parameter.

get_product

ChatGPT
Return one AI-friendly product by ID, MPN or GTIN. Optional fields limits the returned product keys. fields: any of id, mpn, sku, gtin, title, category, url, image_url, gallery_photos, availability, price, price_tiers, material, finish, load_capacity_kg, params, brand, is_variant, data_quality, customs (HS/TARIC codes: hs_code, taric_code_eu, hs_code_us, hs_code_ca, hs_code_gb), documents, material_finish. Default (unset) returns a standard subset without brand/is_variant/data_quality/customs. "material_finish": material and finish name, summary, page url (null if no page), properties (nickel, allergy, corrosion, water, care) and material composition. Use it for nickel-free/allergy questions. An empty-value placeholder ("Not documented", translated) means we do not state it — never "no". A nickel test note covers the finish type, not this product. "documents" lists downloadable PDFs linked to this exact variant (safety data sheet, test report, certificate, manufacturer sheet, manual…) as kind, url and our internal Czech title. "image_url" is the single catalogue shot. "gallery_photos" are the photos the customer sees on the product page (up to five, the same selection as the website): real-life shots of the product in use, and for saddlery tools also technical close-ups. Each has "image_url", "thumbnail_url", a localized "title", and "gallery_page_url" when the photo has its own public page in our gallery — null when it does not, so never invent one. Only this tool returns gallery_photos, documents and material_finish; asking for them in a list result is rejected, because there it would mean an extra query per product. If the MPN is not found, the error includes "suggestions" — real product codes sharing the same prefix (useful when the given MPN is a base code missing its size/variant suffix). PRODUCT_DISCONTINUED means the product existed but is no longer sold. product_id is an alias of id.

get_product_availability_matrix

ChatGPT
Return variant availability in a matrix-like list for axes such as COLOR, WIDTH and DESIGN. Rows are paginated (limit/offset, default limit 50); summary and pagination.total reflect the full filtered set. Every axis must be a parameter of the category — an unknown axis returns PARAM_NOT_FOUND with the valid parameter names in "suggestions" (use get_category_filter_schema to look them up first).

get_product_price

ChatGPT
Return a product price for quantity/currency, with quantity-tier pricing applied automatically. Parameters: mpn — required product MPN/SKU quantity — requested quantity, must be at least 1 (0 or negative returns INVALID_QUANTITY, it is not silently treated as 1); unit_price reflects the tier for this quantity currency — ISO 4217 code, default CZK country — optional ISO 3166-1 alpha-2 code, validated the same way as in get_shipping_methods (unknown code returns COUNTRY_NOT_FOUND). Still informational for the price itself — no country-specific tax/customs adjustment is applied; a note is returned when set. Use get_shipping_methods for country-specific duty/DDP handling instead. customer_group — optional; retail wording (b2c / retail / consumer / private / end customer) is understood and returns the price without a warning. Only B2C pricing is available from this tool today, so wholesale wording (b2b / wholesale / reseller / dealer / distributor / trade / business) returns a warning rather than a B2B price — do not present the returned price as the B2B price in that case. Anything else returns UNSUPPORTED_CUSTOMER_GROUP with the accepted values. incoterm — optional, currently informational only — no DDP/DDU price adjustment is applied; a warning and note are returned when set. lang — output language Check "warnings"/"notes" in the response before answering — they flag when a requested parameter (country/customer_group/incoterm) could not actually be applied to the price shown.

get_shipping_methods

ChatGPT
Return shipping methods, delivery estimates and prices for a destination country. ALWAYS use it for delivery questions instead of guessing. Parameters: country — required, ISO 3166-1 alpha-2 weight_kg — required for prices; if unknown sum get_product weight_kg × qty or ask cart_value — order value as the customer sees it (VAT-inclusive in the EU); sets the free-shipping threshold and GB customs regime; needs currency currency — ISO 4217 code; required for prices carrier — "UPS", "FedEx" (case-insensitive) service_code — exact service ("UPS_EXPRESS_SAVER"); overrides carrier lang — cs/en/de/fr/pl/es/ru; notes[], estimate.note, customs.note stay English — paraphrase them Response semantics: shipping_available=false — we do not DELIVER there. reason="pickup_only": in-person collection only, never "yes, we ship there"; it holds even when a carrier/service filter empties methods[] — call without the filter to list the pickup option. methods[].available=false — cannot be booked: unavailable_reason "no_price_for_weight" (not offered for that weight) or "not_offered_to_country" (the filtered service does not ship there). estimate — delivery_days = transit days from dispatch (working days unless counts_weekends=true); delivery_date (ISO) is authoritative, don't recompute it. estimate.available=false → nothing to quote, see estimate.note (also happens on bookable methods). Not a guarantee: "typically around N working days". dispatch_date — next dispatch day (today if ships_today=true; cutoff Europe/Prague). price — null without weight_kg+currency (price_requires_); amount excludes VAT, amount_incl_vat is what an EU consumer pays (B2B with VAT ID: without); free_shipping.is_free=true → 0. ddp_available=true — duties and taxes paid at checkout, nothing due on delivery; customs.applies / customs.note(_gb) cover customs and fee questions.

list_categories

ChatGPT
Return a flat list of visible categories with their slugs, names and parent slug. Full unfiltered list is ~165 categories — prefer main_only=True for a top-level overview (optionally with include_children=True to embed each main category's direct children inline instead of listing them again as separate flat entries), or parent_slug to list one category's direct children, instead of fetching everything at once. Args: parent_slug: only categories whose parent has this slug (direct children). Mutually exclusive with main_only — main categories never have a parent_slug. main_only: only top-level categories (no parent). query: match on category name, accent- and case-insensitive. When a query matches nothing, the response is {"items": [], "suggestions": [{slug, name}...]} with categories whose name is close to the query (empty when nothing is close, so an empty suggestions list means the category really does not exist). A suggestion is NOT a match — offer it to the customer, do not silently browse it. limit: cap the number of returned categories. When set, the response shape changes to {"items": [...], "pagination": {...has_more, total...}} instead of a bare list. include_children: for main categories in the result, embed their direct children as a "children" list of {slug, name} instead of listing them separately.

search_faq

ChatGPT
Search pethardware.com FAQ in normalized plain text. Parameters: query — optional search text, accent-insensitive and synonym-aware for common support terms lang — output language: cs / en / de / fr / pl / es / ru category_slug — optional FAQ category slug search_mode — normalized / semantic / exact; semantic currently uses normalized matching fields — any of slug, category_slug, category_name, question, answer_text, answer_html, url, updated_at, video_url, score. Default (unset): question, answer_text, url, updated_at, score. limit — maximum number of FAQ items, default 5 answer_format — plain / html / both, default plain Returns JSON with items, scores and source URLs. HTML is omitted by default. Use this to browse or rank multiple candidate FAQ entries yourself. If you just need one ready-to-send answer to a customer's question, call get_customer_answer instead — it resolves the question to a single FAQ entry (or answer=null when none applies).

search_products

ChatGPT
Search products by natural language (inferred colour, width, category, dog weight) plus optional filters. "query" may be omitted when "category_slugs" is given; omitting both returns MISSING_QUERY. Once the category slug is known use get_category_products — more precise. "filters", param_filters, numeric_filters, range_mode and fields work exactly as there (RANGES, LISTS, range_was_expanded). A non-empty "excluded_by_filter" (multi-field alias) is NOT a partial match — judge by "exact_matches"/"fallback_used"; when "filter_url_missing_filters" is set, "filter_url" is WIDER than the result. With exactly one recognized category the response has "filter_url". search_mode: "normalized" (default) or "exact"; near-miss correction of filter values is automatic and reported in applied_filters. Reading the result: - "exact_matches" is conservative: a product is listed only when EVERY explicit requirement (type, material, colour, size, feature such as reflective) is verifiable on it. Empty means "nothing verifiably matches", NOT "we do not stock it": check "alternatives" and "relaxed_filters" first and never present an alternative as exact. - "unsatisfied_constraints" names what emptied "exact_matches" (e.g. ["material=brass"]); prefer it over "relaxed_filters", which describes the fallback and can name a filter that matched. - "unresolved=<word>" in "unsatisfied_constraints" is a near miss of a catalogue value ("blak" → black) with several equally close candidates: ask which was meant, repeat with it in "filters". Plain words block nothing; with no recognized term ("dog food") every word must match. - "unresolved_terms" were not interpreted: check the results answer the question. - A second measurement ("4 mm thick" beside a "20 mm" width) is a hard constraint when some parameter uses that unit, listed in "query_interpretation.hard_constraints .measurement" ("measurement=4 mm" if it emptied the results). It matches ANY parameter, not necessarily thickness — read the dimension off applied_filters.

translate_url

ChatGPT
Translate a pethardware.com URL into all 7 language versions of the site. Supports: - relative paths: /cs/vsuvky/ - absolute URLs: https://www.pethardware.com/cs/vsuvky/ - any language as input: /en/loops/ or /de/schlaufen/ - fragments: /cs/caste-dotazy/#na-zbozi-specham - categories, products, tags, static pages, account pages Parameters: url — relative or absolute pethardware.com URL. A URL on any other host is rejected — this tool never lends our language URLs to a foreign address. format — text / json. Default text preserves legacy clients; json returns structured data. In json format, entity for a product page carries both mpn (the product the path points at) and variant_mpn (the variant selected by ?v=, null when absent or when the code is not a variant of that product). They differ whenever ?v= is present — the page then shows the variant, so quote variant_mpn in that case.

Capabilities

Writes

Example Prompts

Click any prompt to copy 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/pet-hardware"><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/pet-hardware)