Connect your agent.
Three ways in, same results: a web page any browsing agent can read (kaupia.com/search?q=), a JSON API with an OpenAPI spec, and an MCP server for MCP clients — Claude, ChatGPT connectors, Cursor, or an agent you built. No API key during early access.
1. Add the server
{
"mcpServers": {
"kaupia": { "url": "https://api.kaupia.com/mcp" }
}
}Transport: MCP streamable HTTP, stateless, JSON responses. Health check: GET https://api.kaupia.com/healthz. Listed in the official MCP Registry as com.kaupia/kaupia.
2. Tools
| Tool | Input | Returns |
|---|---|---|
search_products | query, country (default US), limit (≤ 50), max_price_usd | Ranked candidates: product_id, title, shop, rating, variants with price, availability and checkout_url |
get_product | product_id | Live variants, price, stock status, options and checkout links. Call before quoting a specific variant. |
create_checkout | product_id, variant_id, quantity | checkout_id and the store’s checkout URL for that variant and quantity |
report_outcome | checkout_id, outcome: placed · delivered · wrong_item · late · refunded · failed | Confirmation. Outcomes decide future ranking. |
shop_scorecard | — | Per‑store delivered and problem counts observed through Kaupia |
3. A typical flow
search_products(query="unscented dog poop bags, 300+ count", max_price_usd=25)
→ choose product_id
get_product(product_id)
→ choose an available variant_id, confirm with the user
create_checkout(product_id, variant_id, quantity=2)
→ open checkout_url with the user
report_outcome(checkout_id, "placed")4. Rules
- Prices are USD in major units (for example
18.99). Stock is live; don’t cache search results. - Get the user’s approval, with the final price, before every purchase.
- Images and listings belong to the store; show them as provided.
- Expected volume above a few hundred calls a day? Tell us first: [email protected]
5. Roadmap
Order placement and payment inside the tool call (no checkout‑link hop), sources beyond Shopify stores, store onboarding, and a buyer guarantee. For agent‑side guidance see For AI agents; the whole site in one file is at llms-full.txt.