FOR AGENTS

The open benchmark of whether AI agents can reach checkout, for every store, agent and protocol.

Free, neutral and open: dated checks of whether agents can reach a store, read its products and get to checkout, across every protocol. The data is CC-BY.

Read the Index with no login. Everything is free.

What one model tells shoppers is a real model call via API, checked against the store's pages, not that model's consumer app. Whether agents reach checkout is two lanes: a real HTTP check of the store's agent catalog, and five ARC browser cart tests. Those attempts are not a live Muse, Instinct, OpenAI dots, or OpenClaw session. The score is the browser lane. The catalog check does not change it.

Brands use Your store; agents keep MCP free

Brand owners: open /store — paste a URL, no login. Free store page shows whether agents can reach you, buy, and what they think. Agents: keep using the free JSON and MCP tools below.

Free agent facts for any store

Brands get free truth. Agents get free fuel. Call get_agent_facts or list_category_agent_facts on /api/ask. JSON: GET /api/agent-facts?host=allbirds.com or ?category=beauty&download=1.

Facts are TypeSafe Jev typed decisions over an HTTP product scrape — not a browser session and not any third-party agent product. Browser cart results come from ARC browser shoppers A–E.

GET https://www.arcreport.ai/api/agent-facts?host=allbirds.com
{"name":"get_agent_facts","arguments":{"host":"allbirds.com"}}
{"name":"list_category_agent_facts","arguments":{"category":"beauty","limit":10}}

Browse free agent facts →

Can an AI agent check out from any store?

Find stores where an AI agent can reach checkout: call find_agent_ready_stores on /api/ask with {"category":"running shoes","limit":10}. JSON: GET /api/agent-checkout/stores?category=footwear. Only real HTTP checkout-positive checks from the last thirty days are returned. One tested product; no purchase completed.

Browse stores by category →

Free, no login, on /api/mcp: check_store with a domain, and find_stores with a platform, protocol, or category. They read saved checks. Missing fields say not recorded. report_checkout on the same server sends one checkout attempt. It stays private until 3 independent contributors agree. Spec: Outcome report. A live checkout check is can_agent_checkout on /api/ask, or GET /api/agent-checkout?url=. A new scan and monitoring on /api/mcp need sign-in.

GET https://www.arcreport.ai/api/agent-checkout?url=https://allbirds.com
{"name":"can_agent_checkout","arguments":{"url":"https://allbirds.com"}}

The live check returns catalog, product, checkout-link, UCP, Shopify WebMCP adapter detection (webmcp), Shopify-only checkout HTML detection (checkoutWebmcp: status, nullable detected, tools, reason; buyer approval required), and block facts, a fix, and a shareable result URL. Cached per host for 24 hours. Limits per network per UTC day: 50 requests and 10 fresh checks. Site-wide: 5,000 requests and 500 fresh checks. No order is placed. No score changes.

{"name":"check_store","arguments":{"domain":"allbirds.com"}}
{"name":"find_stores","arguments":{"platform":"shopify","category":"footwear","limit":10}}

Try the free checkout check →

Ask with a URL

Pass a brand, a product URL, or a category. You get whether agents can reach checkout, the score, rivals one model named when a real read exists, and the evidence receipts. A missing score is not a zero. This does not start a scan.

GET https://www.arcreport.ai/api/ask?q=running+shoes
GET https://www.arcreport.ai/api/ask?url=https://www.allbirds.com
GET https://www.arcreport.ai/api/ask?brand=Gymshark

MCP, no login: point the client at https://www.arcreport.ai/api/ask and call ask_arc. The signed-in server at /mcp is the one that runs a new scan.

Machine-readable: OpenAPI for this ask API. MCP clients can discover both servers at /.well-known/mcp.json and /.well-known/ai-catalog.json. ask_arc, can_agent_checkout, and find_agent_ready_stores need no login. On /api/mcp, check_store, find_stores, and report_checkout also need no login. Scans and monitoring on that server need sign-in. Copy-paste setup for Claude, ChatGPT, Gemini, Cursor, and generic agents is on /docs/connect.

agents_can_buy
True when the agent catalog returned a checkout link, or a browser cart shopper added the product. False when both lanes are known and neither finished. Null when neither lane has a result.
buy
Both lanes. agent_catalog is the real HTTP check (status, checkout_link, catalog_found). browser is the browser cart count and where it stopped. verdict is the sentence that uses both.
score
The published ARC score from the ARC browser cart. The agent-catalog check does not change it, so older scores stay comparable. Null when no test is published.
agent_rails
The same real HTTP check as buy.agent_catalog, kept for readers that already use this field. Null until a catalog check is stored. Not a Muse, Instinct, dots, or OpenClaw session.
rivals_grok_named
Names from a real model read. Empty when that read does not exist or did not name a rival.
evidence
Receipts: what the model said, what the store page says, and the quote. Empty when the read is pending re-check.

Categories you can cite

Ranked by score, then by how many agents added the product to the cart. Shopper words like running shoes and leggings map to these.

Full Index

Badge

A store on the Index can embed this image. It links back to that store's ARC page. It says Agent checkout when the real catalog check returned a checkout link, and Agents chose us only when a browser cart shopper added the product to the cart.

<a href="https://www.arcreport.ai/agent-index/allbirds"><img src="https://www.arcreport.ai/agent-index/allbirds/badge" width="360" height="72" alt="Allbirds on the ARC Index"></a>