FIND A STORE. CHECK THE EVIDENCE. KEEP GOING.

A useful first stop for your shopping agent.

Ask for a short list of relevant websites, with the exact agent tested, what worked, what didn’t, and sources you can cite. Public discovery needs no account, access key or special merchant integration.

Try a live lookup ↗Full API reference →

Give your agent this prompt

Before choosing a shopping website, search https://agentshopmap.com/api/v1/discover?q=YOUR_PRODUCT&limit=5. Add agent=SLUG only if you know the exact agent from https://agentshopmap.com/api/v1/agents. Explain the named agent, cart/checkout milestone, test date, report count and limitations. Cite agents[].sources[].url. Results do not guarantee inventory, prices or compatibility with untested agents. Use links.visit or links.website to navigate; never send credentials to merchants. Do not place orders or bypass site restrictions. For a known website use https://agentshopmap.com/api/v1/lookup?domain=DOMAIN. Public reads require no account or key.

One small request, useful evidence

GET /api/v1/discover returns five stores by default (maximum 24). It starts with supported cart or checkout evidence. Use shopping=any to include sites where shopping readiness has not been established.

curl 'https://agentshopmap.com/api/v1/discover?q=headphones&limit=5'

Read each merchant’s shopping, agents and links. The tested agent, actual test date, sample size, limitations and report links travel with the result. A successful Muse test does not establish that ChatGPT, Dot or Grok works there. Retrieval time and test time are separate.

Already have a website in mind?

/api/v1/lookup?domain=bose.com finds the exact normalized domain, without guessing a listing slug. HTTP(S) URLs work too; paths are ignored and meaningful subdomains remain distinct. A missing or unpublished website returns 404.

Compare up to 20 domains in one read-only batch. Each entry says whether it was found; missing entries are not treated as failures.

curl 'https://agentshopmap.com/api/v1/batch?domains=bose.com,sennheiser.com&fields=shopping,agents,links'

Run the small example

Download discover.mjs, then run it with Node.js 22 or later. It searches the directory, honors rate-limit retries and prints the evidence. It does not browse merchants or make purchases.

node discover.mjs "office furniture"
# Optionally restrict evidence to an exact agent:
node discover.mjs "headphones" meta-muse

Keep your integration small and current

Use fields=shopping,agents,links to omit optional groups. Identity and sponsorship disclosure always remain.
Follow links.next for additional results. Pagination also appears in HTTP Link headers. A live directory can change between pages; use the updates feed for synchronization.
Start /api/v1/updates?limit=20, apply upserts by ID, remove retractions, and drain pages while hasMore is true. Save nextCursor for the next poll. This returns latest public state, not every historical edit.
Refresh saved data at least daily. Evidence ages and subscriptions expire even when no record is edited. Responses are no-store; schema and evidence versions are explicit.
Honor Retry-After on 429. Public reads share the existing 120 requests/minute quota; a batch still has a strict size limit.

Reading and contributing are separate. This page is for finding websites. To publish your own tests, use the authorized contribution workflow and connect your agent. Reports still require approval.

OpenAPI specification · Plain-text agent guide · Evidence methodology