For AI assistants and agents

Shopping assistants can ask BuyWise two things before they recommend something: is this brand what it seems, and what should a shopper know about this listing? Answers come back in a fraction of a second.

What an answer contains

Brands: what kind of brand it is (established, independent maker, store brand, generic white-label), its reputation for quality and where it usually sits on price.
Listings: the same checks the extension runs: generic white-label, quality, price vs typical, fake discounts, review credibility and seller legitimacy.
Every finding carries its confidence ("very likely", "likely", "unsure"). BuyWise never says buy or don't buy; the assistant and the shopper decide.

Connect over MCP

One endpoint for every assistant that supports remote MCP servers: https://api.buywise.fyi/mcp. No sign-up is needed. It offers three read-only tools: check_brands, check_product and check_products.

ChatGPT: Settings → Apps → Advanced → Developer mode, then create an app with the URL above and no authentication.
Claude Code: add the URL with your API key as a header (leave out the header to use the free checks):

claude mcp add --transport http buywise https://api.buywise.fyi/mcp --header "Authorization: Bearer <API key>"

Claude's apps (claude.ai, desktop and mobile) need a BuyWise account connected before the first check.
Muse: ask Muse to add a custom connector for the URL above.
Your own agent (OpenAI Responses API):

{ "type": "mcp", "server_label": "buywise",
  "server_url": "https://api.buywise.fyi/mcp",
  "require_approval": "never" }

Or call the REST API

curl 'https://api.buywise.fyi/agent/v1/brand?name=Anker'

curl https://api.buywise.fyi/agent/v1/products \
  -H 'content-type: application/json' \
  -d '{"products": [{"title": "Anker 737 Power Bank", "brand": "Anker",
       "price": 79.99, "listPrice": 149.99, "rating": 4.6,
       "reviewCount": 8123, "url": "amazon.com"}]}'

Send what the listing shows, with the title exactly as the listing shows it; only title is required. Field names from OpenAI's Agentic Commerce product feed (sale_price, seller_name, star_rating, review_count…) work too. The full schema is at openapi.json and a plain-language guide for models at llms.txt.

Limits and keys

Checks are free. Each brand or listing checked counts as one check, and a daily limit per caller keeps the service fair. Days are UTC.
API keys are made in the BuyWise browser extension's settings. Send one as Authorization: Bearer <key>, over MCP or REST.
No key? Over REST, an agent can pay per check with x402: $0.01 per check in USDC on Base, charged only when the check succeeds. MCP never asks for payment.
Every answer carries quota: plan, used (checks counted today), limit (today's free checks), resetsAt and, on Pro, period (the billing month's start, end, included and used).

Errors

Errors come back as {"ok": false, "error": "<code>", "message": "…"}, with retryAfter in seconds when waiting helps.

Error	Status	Meaning
`bad_request`	400	Malformed request. The message names the first bad field.
`unauthorized`	401	The API key is malformed, unknown or revoked, or this connection needs a BuyWise account connected first.
`payment_required`	402	Pay for the request with x402. Only over REST, and only for callers without an API key.
`blocked_site`	403	The listing's site isn't checked.
`not_found`	404	No such endpoint.
`too_many_items`	413	The batch is larger than the limit.
`quota_exceeded`	429	Today's free checks are used up. They come back at 00:00 UTC.
`allowance_reached`	429	The Pro plan's checks for this billing month are used up.
`throttled`	429	Today's fair-use limit is reached.
`rate_limited`	429	Too many requests this minute.
`budget_exhausted`	503	The service is busy; try again later. A listing check answers 503; a brand check still answers, with its findings empty and `"error": "budget_exhausted"`.
`upstream`	503	A check couldn't be completed. Try again in a moment.

In a batch, an item that can't be checked carries its own error and the rest still succeed.

Privacy

Only a store's domain is used, never the full page address. Banking, health, government, email and sign-in sites are refused, and personal details are removed before anything is checked. No ads, no affiliate links, nothing sold. See the privacy policy.