Product search API for AI agents, with attribution

The call is POST /v1/offers/search on https://api.getclickless.ai. The API reference is the field list. Two fields are required: intent, the product need in the person's words, and ship_to.country. Leave names, emails, phone numbers and street addresses out of intent. Region and postal code on ship_to are optional. limit runs from 1 to 10 and defaults to 3.

The path says "offers" because that is the resource name. The discount the person sees is the deal, in deal.summary, and you show that summary as it arrived.

Each result you might show carries:

• Product title, price in integer cents after the deal, currency, and availability. price_cents is the price after the deal, before tax and shipping.

• Brand name and domain.

• disclosure_text, shown beside the product, word for word, every time you present it.

• sponsored. When it is true, label the result sponsored.

• A fresh attribution_token, and checkout.url, the link the person taps. Approved keys receive both on every result.

Show three products at most, even when you asked for more so you could filter. Those showing rules are part of the creator and developer terms. Quote the price, the deal and the availability as returned. Do not invent a discount, a price or a stock status. To re-check a price later, call GET /v1/offers/{offer_id}. That read has no token. Search again before you show the product to a person, so the sale can be credited.

The API reference, read November 5, 2026, lists 120 requests a minute per key. Past that, the response is 429 with Retry-After set. The OpenAPI description sets a daily search quota of 1,000 searches a day by default.

Keys come after an invite and review. Join the waitlist. The button on For developers opens that list. Self-serve trial keys are coming soon. Production accepts a live key for one agent, sent as Authorization: Bearer plus the key.

The quickstart has the same calls in TypeScript, Python and curl. This is the shape:

const api = "https://api.getclickless.ai";
const headers = {
  Authorization: `Bearer ${process.env.CLICKLESS_API_KEY}`,
  "Content-Type": "application/json",
};

const search = await fetch(`${api}/v1/offers/search`, {
  method: "POST",
  headers,
  body: JSON.stringify({
    intent: "daypack for weekend hikes, with a laptop sleeve",
    ship_to: { country: "US", region: "AZ" },
    limit: 3,
  }),
});
const { offers } = await search.json();
if (offers.length === 0) throw new Error("No products for that intent");

// After the person says yes to this product.
const token = offers[0].attribution_token;
await fetch(`${api}/v1/attribution/approve`, {
  method: "POST",
  headers,
  body: JSON.stringify({ token }),
});
const handoff = await fetch(`${api}/v1/checkout/handoff`, {
  method: "POST",
  headers,
  body: JSON.stringify({ token }),
}).then((response) => response.json());
// Send the person to handoff.url. Pass handoff.discount_code along when it is set.

A shortened example of one result. Example data, fictional brand. The live disclosure_text is whatever the response contains, and that is the line you show.

{
  "offers": [
    {
      "product": {
        "title": "Weekend daypack",
        "price_cents": 17010,
        "currency": "USD",
        "availability": "in_stock"
      },
      "brand": { "name": "Summit Trail Co.", "domain": "summittrail.example" },
      "deal": { "summary": "Deal already reflected in the price" },
      "sponsored": false,
      "disclosure_text": "Heads up: I earn from links I share.",
      "attribution_token": "4fK2pQ9xLm3TzR8vB1nW7c",
      "checkout": {
        "method": "link",
        "url": "https://go.getclickless.ai/t/4fK2pQ9xLm3TzR8vB1nW7c"
      }
    }
  ]
}

Summit Trail Co. is a fictional outdoor brand. The numbers on the screen are example data, not a reported result. The person pays the brand, and the brand ships. Clickless never takes the payment and never holds the goods.

The token is how the sale is traced from the result you showed to the order in the brand's store.

1. Search returns a new attribution token on every result you might show.

2. Show disclosure_text with that product before any link.

3. The person says yes to that specific product. Then call POST /v1/attribution/approve with the token. Do not approve for them, and do not buy on their behalf.

4. Call POST /v1/checkout/handoff with the same token. Give the person url, and discount_code when the response includes one. Do not open, fetch or preview that URL. The brand's checkout charges the card. For a direct brand, the link is the brand's page with the token attached.

5. The store reports the order. Shopify stores report it on their own. Stores not on Shopify send their order export each week to be matched.

6. Without that store report, POST /v1/conversions/report records a claim, not a sale. A claim earns nothing unless the brand's records match it. The earnings page shows the balance while you wait.

One order is credited to one agent. It is not billed twice. A token more than 60 days old has expired, and the API reference says to search again. An order that did not start from an ad is credited within 60 days of the product being shown.

Developers are paid from the commission Clickless receives. Payouts are rolling out. Until they switch on, earnings wait in the balance.

A developer's agent sees direct brands' products. Affiliate network catalogs follow as each network approves sharing them.

Age-gated products come back when you send constraints.age_verified set to true, and you send that after you have verified the person's age. When regulated_tier is true, stay with the brand's own product words and add no claim of your own. When no_health_claims is true, the description is withheld. Do not fill the gap.

If you send user_context.user_ref, use a pseudonymous id. Clickless hashes it with a key unique to the agent and does not store the raw value.

Drafting a store's Product Expert from its URL through the API is coming soon. Leave it out of a build until the docs say it is on.

The Model Context Protocol (revision 2025-11-25, published November 25, 2025) is an open protocol for connecting tools to an LLM application. A server exposes tools a client can call.

Clickless's MCP server is the same service as the REST API, at https://mcp.getclickless.ai/mcp, with the same live key in the Authorization header. search_offers maps to the search call and returns three results at most. approve and checkout_handoff map to the two calls above. get_offer re-checks price, deal and availability and returns no token.

Use REST when you want to filter in your own code, then show three. Use MCP when the agent already speaks the protocol and you want each tool's description to carry the showing rules: disclosure text on every result, three at most, no invented prices, and approval after the person says yes.

IntentLink and Channel3 also publish product search APIs. The lines below are from their own pages, read November 5, 2026. Re-check a page before you depend on a field name. Clickless is not affiliated with either company, or with Amazon or Walmart.

When to choose IntentLink. Their documentation, last updated July 24, 2026, covers shopping plus hotels, flights and tickets, labels the API beta, and says a typical search is under one second at the 90th percentile. Choose them when the agent needs travel or events as well as products, after you read the current parameters.

When to choose Channel3. Their developers page, accessed November 5, 2026, says the catalog is 100 million products across 25,000 retailers, with picture search, monetizable links and a median search latency of 800 milliseconds, and marks programmatic checkout as coming soon. Choose them when you want that wider catalog, after you read the commission terms on their site.

When to choose Clickless. Choose Clickless when the person should finish on the brand's own checkout, see a disclosure line, and have the sale credited to one agent. Developers are paid from the commission Clickless receives. Payouts are rolling out. Clickless does not control which products another company's model names. For the earning side of recommendations, see the affiliate program for AI agents.

Walk one real question through search, approval and handoff. Confirm the disclosure text is visible, three products at most are shown, approval waits for a yes, the agent does not fetch the handoff link, intent has no personal data, and a saved product is searched again so the token is new.

The API reference lists the errors: a missing field, a bad key, a token that was never approved, an expired token, or a checkout that cannot take the order. Join the waitlist at getclickless.ai/agent-developers. Put the showing-offers rules in the agent's instructions.