| 开发者 | carthub |
|---|---|
| 更新时间 | 2026年7月14日 18:40 |
| PHP版本: | 8.0 及以上 |
| WordPress版本: | 7.0 |
| 版权: | GPLv2 or later |
| 版权网址: | 版权信息 |
/discovery endpoint — the single entry point any agent needs to understand your catalog/catalog/search, /catalog/products, /catalog/product/{id}, /catalog/categories endpoints/wp-json/kalicart/v1/mcp (JSON-RPC 2.0) so MCP-capable agents can call the catalog as tools — same data as the REST endpoints, no API keyproducts_url and search_url_template fields so agents can navigate without constructing URLs manually<link rel="kalicart-agent"> in your site <head> for agent auto-discoveryAllow: /wp-json/kalicart/ in robots.txt/kalicart-sitemap.xml linked from the WP sitemap indexpa_gender attribute, category paths, tags, product name (multilingual keywords: IT/EN/FR/DE/ES)pa_color/pa_colore and product metadatapa_size/pa_taglia, type auto-detected (clothing S/M/L, numeric EU, shoes EU half-sizes)/wp-json/kalicart/v1/mcp (JSON-RPC 2.0 over HTTP POST). MCP-capable agents and assistants connect to it and call the catalog as tools: search_products, list_products, get_product, list_categories, get_meta. It is read-only and needs no authentication, exactly like the public REST endpoints — it adds a second transport, not new data. No LLM, no external calls. Checkout and payment are never exposed over MCP.
/wp-content/plugins/kalicart-bridge/yourdomain.com/wp-json/kalicart/v1/discoveryyourdomain.com/wp-json/kalicart/v1/mcphttps://bridge.kalicart.com/docs/.The feed format is region-neutral. Merchant eligibility and shopping-surface availability are determined by OpenAI and may vary by market.
No, by design. The feed is discovery-only (is_eligible_checkout=false): customers buy on your storefront and you remain merchant of record. This matches OpenAI's own shift toward merchant-owned checkout.
Not unless you choose to. By default the plugin runs entirely on your server and sends nothing externally. The only exception is the optional Federated Catalog feature: if you explicitly activate it, the plugin sends your store's public URL (and nothing else) to KaliCart Global so your public catalog can be included in federated agent search. No customer, order, or private data is ever sent. See "External services" below, and the privacy notice at https://bridge.kalicart.com/privacy/.
No. This plugin is fully standalone and free.
The Federated Catalog is an optional discovery network operated by KaliCart Global. The Bridge makes your catalog readable by agents that already reach your domain; the Federated Catalog makes it discoverable by agents that do not know your store yet. It is opt-in, separate from the local Bridge endpoints, and revocable at any time.
There are two discovery paths. Local signals — .well-known files, robots.txt entries, the rel="kalicart-agent" head link and the badge — help an agent that lands on your own domain find the Bridge. They do not feed the federated index by themselves. The federated index is a separate cross-merchant index: after you activate it, KaliCart Global reads your already-public /discovery and /catalog/* endpoints and lets agents search across participating stores.
End to end: from WP Admin → KaliCart Bridge you activate the Federated Catalog; the plugin sends only your public site URL; KaliCart Global pulls the public catalog read-only; matching results route agents back to your store. Authoritative price, availability and checkout stay on your WooCommerce site, served live by your Bridge. If you revoke consent, your catalog leaves federated results while direct agent access to your store remains active.
Public endpoints (discovery, search, products, categories) are accessible without authentication — same as the WooCommerce REST API public surfaces. The /health endpoint requires manage_woocommerce capability.
Yes — all three signals (badge, robots.txt, sitemap) can be toggled individually in the KaliCart settings tab.
5 minutes. You can force a refresh via the "Refresh analysis" button or the ?force=true query parameter on the /health endpoint.
The catalog can be consumed two ways. Any agent can call the plain REST endpoints under /wp-json/kalicart/v1/catalog/ directly. MCP-capable clients can instead connect to the MCP server at /wp-json/kalicart/v1/mcp (JSON-RPC 2.0): the assistant connects to that URL — no API key — and gains the catalog tools (search_products, get_product, list_categories, get_meta, list_products). In clients that support remote MCP connectors, add it as a custom connector pointing to that URL.
Yes, when the chatbot service can ingest URLs, API documents or external knowledge sources. Give it the discovery URL first: https://yourdomain.com/wp-json/kalicart/v1/discovery. From there it can find the catalog endpoints, the OpenAPI description and the MCP endpoint. If the service supports live REST or MCP calls, it can query current catalog data directly. If it only imports a knowledge base, it may create a snapshot: useful for product answers, but final price, stock, coupons, shipping and checkout should still be verified on the merchant site.
The benefit is practical: chatbot builders can read a structured, machine-readable WooCommerce catalog instead of scraping product pages. The Bridge does not add an LLM to your site and does not decide how the chatbot works; it supplies cleaner catalog data for tools that can consume it.
, non-breaking space) are no longer dropped from the ChatGPT feed. The product name is used as a fallback, and entity-only descriptions no longer leak into the feed. Plain-text extraction now decodes HTML entities.fields query parameter on /catalog/products and /catalog/search. fields=summary returns a slim per-item projection — id, sku, name, url, price.current, price.display, stock.in_stock, categories, type and updated_at — so an agent can triage a large catalog cheaply and open /catalog/product/{id} in full only for the candidates worth pursuing. fields=full remains the default, so existing consumers are unaffected. On a full page the summary payload is roughly an order of magnitude smaller than full, and on browse and search without post-filters the response is also assembled with far fewer database queries because the heavy per-product work (attribute terms, images, gender/colour inference, shipping, variants) is skippedq is the only text-search parameter, per_page is the result-count parameter, /catalog/search is for text search, and /catalog/products is for browsing/listing. Invalid aliases such as query, limit, or q on /catalog/products now return a guided 400 response with the corrected endpoint and suggested_url