# ARC — Agentic Readiness Check > The open benchmark of whether AI agents can reach checkout, for every store, agent and protocol. ARC is free, neutral and open. It publishes dated, CC-BY-4.0 measurements of whether AI agents can reach a store, read its products and get to checkout, across every protocol (UCP, ACP, PAP, WebMCP, MCP, agent payments). It takes no payment from any protocol, platform or model company. Anything ARC has not measured is labelled "not measured", never zero. Start here: - Open dataset (CC-BY-4.0, weekly CSV + Parquet): https://www.arcreport.ai/data - Protocol Matrix (Declared / Valid / Usable / Transacts per protocol): https://www.arcreport.ai/protocols · JSON: https://www.arcreport.ai/api/protocols - Open agent leaderboard (cart success per agent path on ARC-100, a fixed public set of 100 real stores, with n and 95% intervals): https://www.arcreport.ai/leaderboard · JSON: https://www.arcreport.ai/api/leaderboard · MCP agent_leaderboard - ARC-100 v1 test set: https://www.arcreport.ai/data/arc-100/v1/arc-100.csv · Submit your agent's run (unverified until ARC re-runs a sample): https://www.arcreport.ai/leaderboard/spec · POST https://www.arcreport.ai/api/leaderboard/submit · MCP submit_leaderboard_run - Check any store instantly (free, no login): https://www.arcreport.ai/agent-checkout - Submit stores (up to 50, paste or CSV): https://www.arcreport.ai/agent-checkout/submit · POST https://www.arcreport.ai/api/stores/submit · MCP submit_store - Claim your store card (free: sign in with an email magic link, then verify by DNS TXT or HTTPS file): https://www.arcreport.ai/store/{host}/claim - Methodology: https://www.arcreport.ai/methodology · Method changelog: https://www.arcreport.ai/methodology/changelog · Citation: https://www.arcreport.ai/CITATION.cff - Free MCP for agents, no login: https://www.arcreport.ai/api/ask - Developers (free tools, no key): https://www.arcreport.ai/developers · UCP profile validator (published UCP schema + spec rules): https://www.arcreport.ai/developers/ucp-validator · GET https://www.arcreport.ai/api/ucp/validate?url= · MCP validate_ucp_profile - MCP tools reference (arguments, curl/JS/Python and REST twin for every public tool): https://www.arcreport.ai/developers/mcp · OpenAPI 3.1 for every public endpoint: https://www.arcreport.ai/openapi.json - arc-check CLI and CI step (fails when a store's UCP profile or WebMCP breaks): curl -fsSLO https://www.arcreport.ai/arc-check.mjs && node arc-check.mjs yourstore.com · OpenAPI: https://www.arcreport.ai/openapi.json Live directory size: 3,329 listed stores (curated, Index and submitted; new ones are queued for their first check), counted at 2026-10-11 13:58 UTC. Live pages show live counts; dataset files are dated snapshots. The Index's five browser shoppers are ARC personas, labelled ARC browser shopper A to E. They run on ARC-chosen models and are not the consumer agents of similar style. Their runs stop before payment; nothing is ever bought. One AI model's read-back (currently Grok via API, not the consumer Grok app) is checked against store pages and shown as one model's view, not a consensus. Whether agents reach checkout is two lanes: the agent-catalog lane is a real HTTP check (catalog tools and a valid Universal Commerce Protocol profile at /.well-known/ucp); the browser lane is five ARC browser cart tests. The catalog check spends no model call and does not change the score. An HTML page or a bot block is not a profile. ARC is not affiliated with Meta, Spear Street Technology, SpaceXAI, OpenAI, or the OpenClaw Foundation. A published Index score is the browser lane. If the store returns a bot challenge, the result says the store blocks automated agents and includes the HTTP evidence. If ARC could not finish the test, the page says retest pending. A missing score is not a zero. ## Free brand hub (Your store) — no login Free store page shows whether agents can reach you, buy, and what they think. Paste a store URL. See checkout, WebMCP, agent facts, Index score, read-back mismatches (one AI model), and a simple history in one place. An empty hub automatically runs the free HTTP checkout check once; existing cache and rate limits apply. No email or browser scan. /store/{host} includes the free Agent Reachability Matrix: discover, buy / checkout, and allow vs block risk. Every cell labels measured, inferred, unknown, or public report evidence. Rails checks and Index browser cart tests are not live consumer-agent sessions. Hub: https://www.arcreport.ai/store Example: https://www.arcreport.ai/store/allbirds.com No paywall. Sign-in optional. Public agent MCP stays free at /api/ask. Personal Agent Protocol (PAP, announced Oct 2026 by Meta + Sierra with partners): open standard for how personal agents interact with businesses. ARC watches for discovery markers on /store/{host} and ties them to the two-lane question (can agents discover you / buy from you). Until v0.1, status is "watching for PAP v0.1" unless a sensible candidate marker appears — ARC does not invent protocol support and is not affiliated with Meta or Sierra. Announcement: https://sierra.ai/blog/introducing-personal-agent-protocol ## Free agent facts (no login) Brands get free truth. Agents get free fuel. Hub: https://www.arcreport.ai/agent-facts JSON: https://www.arcreport.ai/api/agent-facts?host=allbirds.com Download: https://www.arcreport.ai/api/agent-facts?download=1 MCP on /api/ask: get_agent_facts({"host":"allbirds.com"}), list_category_agent_facts({"category":"beauty","limit":10}) Facts are TypeSafe Jev typed decisions (typesafe/jev-1.13 via OpenRouter Decisions API) over HTTP product scrapes. Hard daily Jev spend cap (default $1). Not a live consumer-agent session. Browser Index cart tests are unchanged. ## Ask ARC (no login) Any agent can read the Index. This does not start a scan and does not require an account. - Docs: https://www.arcreport.ai/for-agents - OpenAPI: https://www.arcreport.ai/openapi.json - MCP discovery: https://www.arcreport.ai/.well-known/mcp.json and https://www.arcreport.ai/.well-known/ai-catalog.json - 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 - GET https://www.arcreport.ai/api/ask?category=footwear - MCP tool `ask_arc` on https://www.arcreport.ai/api/ask (no login). The signed-in server that runs a new scan is https://www.arcreport.ai/api/mcp The JSON returns, per match: `verdict`, `agents_can_buy`, `buy` (agent_catalog and browser), `score`, `could_buy`, `shoppers`, `agent_rails` (the same live HTTP check as buy.agent_catalog), `rivals_grok_named` (only when a real Grok read exists), and `evidence` receipts (what Grok said, what the store says, the quote). `agents_can_buy` is true when the catalog returned a checkout link or a browser cart shopper added the product. `could_buy` stays the browser count. `agent_rails` is null until a catalog check has run. It does not change `score`. Shopper words such as running shoes and leggings map to the live Index categories. State of agent checkout (live counts, no login): https://www.arcreport.ai/agent-checkout ## Category Index Ranked by score, then by how many browser cart attempts added the product to the cart. - Apparel: https://www.arcreport.ai/agent-index/category/apparel - Footwear: https://www.arcreport.ai/agent-index/category/footwear - Beauty: https://www.arcreport.ai/agent-index/category/beauty - Supplements: https://www.arcreport.ai/agent-index/category/supplements - Home: https://www.arcreport.ai/agent-index/category/home - Food & drink: https://www.arcreport.ai/agent-index/category/food-drink ## Badge A store on the Index can embed `https://www.arcreport.ai/agent-index/{slug}/badge`. 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. ## Site - Home and free scan: https://www.arcreport.ai/ — Paste a product URL. The result leads with both lanes: the real agent-catalog check, and where the five browser cart shoppers stopped. You get the score, the failure, and the fix, free. The catalog check does not change the score. - Index: https://www.arcreport.ai/agent-index — Where can personal agents reach checkout? Each row shows browser score and a real agent-checkout yes or no. - State of agent checkout: https://www.arcreport.ai/agent-checkout — Counts from stored checks: catalog, product found, checkout link, blocks, and stores with no usable rails. - Your store hub: https://www.arcreport.ai/store — Free brand dashboard for any URL - Agent facts: https://www.arcreport.ai/agent-facts — Free agent facts for any store - For agents: https://www.arcreport.ai/for-agents - Connect your agent (signed-in scans): https://www.arcreport.ai/mcp - AI crawler directory: https://www.arcreport.ai/ai-crawlers - Shoppers: https://www.arcreport.ai/shoppers — One page per ARC browser shopper persona (A to E). Each page counts published ARC Index stores where that shopper's cart pass added the product to the cart, and groups the stops that were recorded. These pages are the browser lane, not a live run of those products and not the agent-catalog check. The catalog check is on the store page, in /api/ask, and at /agent-checkout. A retest pending row is not a score and is not counted. Nothing on these pages is estimated. - ARC browser shopper A: https://www.arcreport.ai/shoppers/muse - ARC browser shopper B: https://www.arcreport.ai/shoppers/instinct - ARC browser shopper C: https://www.arcreport.ai/shoppers/grok - ARC browser shopper D: https://www.arcreport.ai/shoppers/openai-dots - ARC browser shopper E: https://www.arcreport.ai/shoppers/openclaw (URL paths keep the legacy alias ids.) - Methodology: https://www.arcreport.ai/methodology — How the browser cart test, the score, the agent-catalog check, the Agent Reachability Matrix (measured, inferred, unknown, public report), and the PAP watch work. - Blog & reports: https://www.arcreport.ai/blog — Reports from ARC. Published posts are listed under Blog / Reports below. - Contact: https://www.arcreport.ai/contact - Privacy: https://www.arcreport.ai/privacy - Terms: https://www.arcreport.ai/terms A pitch report lives at /pitch/{id}. The id is unguessable and is not the scan id. The page is noindex. It is not a public Index score. Agency name and an optional logo URL or text mark are stored on that pitch. An email is never stored there. Older /report/{id}/pitch links redirect to it. ## Free instant checkout check — any public store GET https://www.arcreport.ai/api/agent-checkout?url=https://allbirds.com Public MCP at /api/ask: can_agent_checkout({"url":"https://allbirds.com"}). No login. JSON: GET /api/agent-checkout?url=. The /api/mcp server requires OAuth for every request. Ask: "Can an AI agent buy from this store?" or "Can personal agents check out on this Shopify store?" The free HTTP check also detects the Shopify WebMCP storefront adapter for browser agents (webmcp boolean), not a live browser-agent session. Returns catalog search answered, product found, checkout link returned (boolean only), UCP profile, webmcp adapter detection, block status, one fix, checkedAt, cached and resultUrl. Homepage checks test one catalog product. Product checks test the supplied product. Cached per host for 24 hours, including failures; the returned productUrl identifies the tested product. No model calls, orders, email, or score changes. Other platforms can honestly return no agent catalog; that does not prove every agent cannot buy. 50 requests and 10 fresh checks per network per UTC day; site-wide 5,000 requests and 500 fresh checks. 429 means capacity or a host check in progress. JSON errors use 400 or 503. Human form: https://www.arcreport.ai/agent-checkout Shareable stored results: https://www.arcreport.ai/agent-checkout/{host} ## Where can my agent buy? Directory: https://www.arcreport.ai/agent-checkout/stores — 3,329 direct consumer storefronts (curated, Index and submitted) across apparel, footwear, beauty, home, food-drink, supplements, outdoor, kids, pets, electronics and jewelry. Anonymous MCP /api/ask: find_agent_ready_stores({"category":"running shoes","limit":10}) or {"q":"skincare"}. JSON: GET /api/agent-checkout/stores?category=footwear&limit=10. Limit 1–50. Reads share the checkout API caps: 50/network/day and 5,000/site/day (UTC). Returns stores (including webmcp adapter detection), verdict, checkedAt and arcUrl only for real HTTP checkout-positive results within seven days. Category matching is not product inventory search. A checkout link for one catalog product is not a completed purchase or a promise about another item. No browser scan or model call. Checkout links are never stored. Empty results mean no recent positive checks, not that no store supports agents. Unchecked stores are pending. Timeouts are inconclusive. No agent catalog found is not a block. Blocks are reported only when observed. The full score uses browser cart shoppers; HTTP checks do not change it. The browser lane uses ARC browser shopper personas A to E. - /badge/{host}.svg (also /store/{host}/badge and /store/{host}/badge.svg): free embeddable "Agent-ready, checked by ARC" SVG from saved agent-catalog evidence, labelled Measured, Inferred or Not measured. Says Agent-ready only when the catalog returned the tested product and a checkout link. Embed it linking back to /store/{host}; cached about an hour; never starts a check. ## Agent commerce adoption tracker Public dated counts, category rates, daily UTC history and changes from saved append-only HTTP checks: https://www.arcreport.ai/agent-checkout/adoption JSON: https://www.arcreport.ai/api/agent-checkout/adoption Anonymous MCP on /api/ask: agent_commerce_adoption({}) returns current counts, categories and byPlatform (homepage fingerprints, measured denominators; Not detected includes headless/custom stores). Reads only; no store requests or model calls. Rates use the latest completed check per host. Failures are reported separately. Category percentages require at least 10 completed hosts. WebMCP is HTML adapter detection, not a live agent session. The sample is ARC’s curated list, not all e-commerce. ## Free research pages (CC-BY-4.0) Dated, citable summaries generated hourly from live checks, each with charts, n per group, evidence labels, a cite block and JSON/CSV downloads: - Agent readiness by commerce platform: https://www.arcreport.ai/insights/agent-readiness-by-platform (JSON /api/insights/agent-readiness-by-platform.json, CSV .csv) - Agent readiness by market (US, UK, Ireland, rest of Europe, Canada, Australia): https://www.arcreport.ai/insights/agent-readiness-by-country - Agent readiness by store category: https://www.arcreport.ai/insights/agent-readiness-by-category - Return and shipping policy clarity for AI agents: https://www.arcreport.ai/insights/return-and-shipping-policy-clarity Index: https://www.arcreport.ai/insights · https://www.arcreport.ai/api/insights/index.json ## Open dataset (CC-BY-4.0) Weekly dated snapshots of ARC's measurements, free to download: https://www.arcreport.ai/data Files: stores (Index results), rails_checks (WebMCP, MCP, UCP, catalog, cart link per store) adoption_history (weekly counts) and protocol_matrix (stage counts per protocol), as CSV and Parquet at https://www.arcreport.ai/data/latest/ with a manifest at https://www.arcreport.ai/data/latest/manifest.json. Dataset card: https://www.arcreport.ai/data/README.md Anonymous MCP on /api/ask: dataset_info({}) returns the latest URLs, row counts and licence. Licence CC-BY-4.0, attribution: ARC, arcreport.ai. Scores, signals, URLs and dates only; no page text or personal data. ## Grow the open dataset: check or submit stores (free, no login) Check any store instantly: https://www.arcreport.ai/agent-checkout (permalink per host at /agent-checkout/ with badge embed and cite block). Every checked store joins the open directory after validation (live HTTPS, not parked or closed, not redirecting to another brand). Submit up to 50 stores (paste or CSV): https://www.arcreport.ai/agent-checkout/submit · JSON: POST https://www.arcreport.ai/api/stores/submit with {"hosts": [...], "category"?} or {"text": "..."} or {"url": "..."} Anonymous MCP on /api/ask: submit_store({url} | {hosts, category?}). Accepted stores are queued for plain-HTTP checks (UCP, WebMCP, MCP, policies, product data) and appear in rails_checks with source=submitted. No model calls; rate limited per caller. ## Method changelog and citation Versioned public method changelog: https://www.arcreport.ai/methodology/changelog (JSON: https://www.arcreport.ai/api/methodology/changelog). Current method: arc-method-2026.10. CITATION.cff: https://www.arcreport.ai/CITATION.cff The Index's five browser shoppers are ARC personas labelled "ARC browser shopper A" to "E" (dataset columns persona_a…persona_e). They are not named consumer agents. Legacy alias ids (muse, instinct, grok, openai-dots, openclaw) appear only as a documented alias field (legacy_alias_id). ## Protocol Matrix (caniuse for agentic commerce) Free, neutral matrix across UCP, ACP, PAP, WebMCP, storefront MCP / MCP discovery, and Visa Trusted Agent Protocol / Mastercard Agent Pay: https://www.arcreport.ai/protocols (one page per protocol at /protocols/: ucp, acp, pap, webmcp, mcp, payments) Four stages per protocol: Declared, Valid, Usable, Transacts, each with count, denominator, percentage and an evidence label (Measured / Inferred / Not measured). Not measured means ARC has no saved evidence, not zero adoption. JSON: https://www.arcreport.ai/api/protocols · CSV: https://www.arcreport.ai/data/latest/protocol_matrix.csv Anonymous MCP on /api/ask: protocol_matrix({protocol?}) returns the same. Saved checks only; no store requests or model calls. CC-BY-4.0, attribution: ARC, arcreport.ai. ## Free product data check product_data_check(host or url) on /api/ask reads the latest saved product data from the page’s structured data. JSON: GET /api/agent-checkout/product-data?host=allbirds.com. No store fetch or model call on reads. JSON-LD from server HTML only; presence does not verify accuracy. Null means measured on the next check. Store details: /store/{host}#product-data. Includes saved return and shipping policy findings from the store’s pages and structured data; presence does not verify terms. Google agentic checkout readiness: google_checkout_readiness(host) on /api/ask (free, no login) or /api/mcp (OAuth) reads saved public UCP, policy, support contact, and product identifier checks. product_data_check and its JSON endpoint include googleReadiness. Missing history is measured on the next check. native_commerce(checkout_eligibility) cannot be checked from outside; set it in the Merchant Center feed. This is not Google approval or certification. See Google Merchant Center help on UCP / agentic checkout. Counts and measured denominators: /api/agent-checkout/adoption, including categories. Store cards at /store/{host} can be claimed free at /store/{host}/claim using DNS TXT or an HTTPS file. Brand claims and statements never change ARC’s measured scores. Store hub JSON: /api/store?host={host}; MCP: get_store_hub(host). Free AI agent robots.txt access: saved home/product-path rules by answer/search/training agent on /store/{host}, can_agent_checkout, store_card and product_data_check; measured latest-completed-host denominators and blocked counts on /api/agent-checkout/adoption and agent_commerce_adoption. Robots policy is not observed fetch access. ## Cost ARC is a free, open research project. The store hub, the Index, product scans, the datasets and the APIs are free. There are no paid plans. ## Blog / Reports Index: https://www.arcreport.ai/blog — Reports from ARC on whether personal agents can find, choose, and buy from stores. Methodology: https://www.arcreport.ai/methodology - State of agent buyability: October 2026 (2026-10-06): https://www.arcreport.ai/blog/state-of-agent-buyability-oct-2026 — ARC's first Index data report. Of 83 stores scored in October 2026, 31 had all five ARC browser cart shoppers add the product to the cart, 21 had none, and 14 of those 21 still returned an agent checkout link over HTTP. - How ARC tests whether agents can buy from your store (2026-10-06): https://www.arcreport.ai/blog/how-arc-tests — ARC tests agent buyability in two lanes: a live HTTP check of your agent catalog and five automated browser cart attempts that show the exact step where checkout stops. - What is agent buyability? (2026-10-06): https://www.arcreport.ai/blog/what-is-agent-buyability — Agent buyability is how easily a personal AI agent can reach your online store, understand your products, and complete a purchase for a shopper. ARC grades it for free.