SuggestAPISuggestAPI

Google ARD · Agentic Resource Discovery

Google Agentic Resource Discovery (ARD).

SuggestAPI implements Google’s Agentic Resource Discovery specification — publishing a live capability catalog so agents can find commerce search, comparison, and checkout handoff without hardcoded tools.

Sign upRead Google’s ARD announcementView live ARD catalogAgent reference
Abstract diagram of agentic resource discovery: a central capability catalog connected to APIs, tools, and commerce search surfaces.

Agent Search Layer

Publish once at /.well-known/ai-catalog.json. Agents and registries discover SuggestAPI commerce capabilities through an open standard.

Discovery, not execution

Google announced Agentic Resource Discovery (ARD) as an open specification for publishing, discovering, and verifying AI capabilities across the web. It helps an agent answer what exists and where to find it — then the selected API or protocol handles runtime auth and interaction. Read the Google Developers Blog announcement or the ARD specification (v0.9 draft).

Without ARD

  • Developers hardcode tool descriptions into each agent
  • Agents are limited to tools configured in advance
  • Every platform invents its own discovery metadata

With ARD

  • Providers publish capabilities in a common catalog format
  • Registries can index capabilities for dynamic discovery
  • Catalogs can include domain-anchored identity and trust metadata

What SuggestAPI publishes

The Knowledge Gateway hosts an ARD-compatible catalog with domain-anchored host identity and structured entries for each resource surface.

Live catalog

https://agent.suggestapi.com/.well-known/ai-catalog.json

Inspect the machine-readable ai-catalog.json, or read the ARD specification (v0.9 draft).

SuggestAPI Agent Registry

Discover commerce search, comparison, recommendation, and handoff capabilities.

Tenant Search Surface

Tenant-scoped product and content search through a documented API surface.

Agent Search Orchestration

Higher-level retrieval, comparison, recommendation, and policy-aware answers.

Agent Cart Handoff

A safe handoff plan with merchant-controlled product, cart, and checkout resources.

Guided Checkout Handoff

Guidance that routes a selected product toward the merchant’s existing checkout.

How agents discover SuggestAPI

An Agent Search Layer that connects intent to live catalog capabilities — without coupling the agent to a specific search engine.

Step 1

Publish

SuggestAPI exposes its ARD catalog from the standard well-known path on the Knowledge Gateway.

Step 2

Discover

An agent fetches the catalog directly, or finds entries through an ARD registry that crawls published catalogs.

Step 3

Select

Descriptions, capability labels, and representative queries help match the user’s intent to the right surface.

Step 4

Connect

The agent uses the resource’s native API or protocol. SuggestAPI routes to the tenant search source and returns structured results.

Merchant control stays intact

Making a capability discoverable does not hand checkout to an agent. SuggestAPI’s cart and checkout resources are handoff surfaces — they prepare links, they do not place orders or process payments.

Keep your search stack

Algolia, Typesense, Elasticsearch, Meilisearch, or a SuggestAPI index stays the source of truth. See Bring Your Own Search.

Structured product data

Assistants search, compare, and recommend from live catalog information — not brittle HTML scrapes or stale model knowledge.

Checkout stays yours

Product discovery becomes agent-ready while your storefront and commerce platform control cart, checkout, and payment.

Agent reference

Request, response, auth, and tool calls

ARD is discovery only. Fetch the catalog with no credentials, pick a resource, then call that resource’s native API. Auth and execution never live in ai-catalog.json. The live catalog is the source of truth; examples below show the contract agents should expect.

1. Discover (no auth)

GET the well-known catalog. Do not send API keys on this request.

GET https://agent.suggestapi.com/.well-known/ai-catalog.json HTTP/1.1
Accept: application/json

Also load https://agent.suggestapi.com/openapi.yaml and MCP discovery when the catalog points at those artifacts.

Catalog shape

Treat field names as ARD 0.9 draft. Read resources[] for id, type, url, and capabilities.

{
  "ardVersion": "0.9",
  "host": {
    "id": "agent.suggestapi.com",
    "url": "https://agent.suggestapi.com"
  },
  "resources": [
    {
      "id": "suggestapi.agent.search",
      "type": "http",
      "name": "Tenant catalog search",
      "url": "https://agent.suggestapi.com/v1/agent/search",
      "capabilities": ["search", "commerce.discovery"]
    }
  ]
}

2. Authenticate

  • Catalog fetch: anonymous.
  • Knowledge Gateway search, compare, recommend: Authorization: Bearer $SUGGESTAPI_KEY.
  • Search acceleration API (autocomplete / query): x-api-key or Authorization: Bearer.
  • Never send Algolia, Typesense, or Elastic credentials. SuggestAPI injects backend keys server-side.
  • Cart and checkout resources are handoff URLs. They do not accept order or payment payloads through SuggestAPI.

3. Example tool call

After selecting the search resource from the catalog, POST a tenant-scoped query.

POST https://agent.suggestapi.com/v1/agent/search HTTP/1.1
Authorization: Bearer $SUGGESTAPI_KEY
Content-Type: application/json

{
  "tenant": "demo.suggestapi.com",
  "query": "waterproof hiking boots",
  "limit": 8
}

Example response

Normalized product objects plus a merchant-controlled checkout handoff. Do not invent SKUs; use live tenant data.

{
  "query": "waterproof hiking boots",
  "results": [
    {
      "id": "sku_1842",
      "title": "TrailGrip Waterproof Hiking Boot",
      "url": "https://demo.suggestapi.com/products/trailgrip",
      "price": { "amount": 18900, "currency": "USD" },
      "availability": "in_stock"
    }
  ],
  "handoff": {
    "checkout": "https://demo.suggestapi.com/cart"
  }
}

Storefront autocomplete

Human typeahead and agent prefix probes use the acceleration API, not the ARD catalog itself. See Bring Your Own Search for adapters and response fields.

GET https://api.suggestapi.com/v1/autocomplete?index=products&query=boot HTTP/1.1
x-api-key: $SUGGESTAPI_KEY
Accept: application/json

ARD FAQ

Short answers for teams evaluating agent discovery standards.

Is ARD a runtime protocol?

No. ARD is a discovery specification. It does not replace OpenAPI, MCP, A2A, or another runtime. Authentication and execution stay with the selected resource.

Do merchants publish their own ai-catalog.json?

Not necessarily. SuggestAPI publishes platform capabilities from the Knowledge Gateway. Merchants can optionally publish a first-party catalog on their own domain.

Is the specification finalized?

Not yet. The current ARD specification is version 0.9 and labeled as a draft. Expect schema evolution as the open project matures.

Where can I read more?

See the announcement blog post, Google’s ARD announcement, and the how it works overview.

How It Works

Agentic commerce without a replatform

Connect your catalog and keep the search engine you already run.

Bring Your Own Search

Adapters, schema, and trust boundaries

Engineering reference for routing Algolia, Typesense, Elastic, or Meilisearch.

Blog

Google ARD Reddit

Short answers for what Agentic Resource Discovery is — and what it is not.

Ready to add agentic commerce to your storefront?

Sign up to connect your catalog. SuggestAPI becomes The Discovery Layer for AI Agents — with an Agent Search Layer that publishes ARD-compatible capabilities while your checkout stays on your rails.

Sign upRead the announcement