A UCP Score is a 0–100 grade measuring how agent-ready a store is. It combines Agent Discovery (30%), UCP Conformance (40%), and Capability Coverage (30%). Scores map to A/B/C/D/F grades. Used by AI shopping agents like Google AI Mode, ChatGPT, and Claude to decide which stores to surface, recommend, and transact with.

How is the UCP Score calculated?

Each category is rated 0–100 then weighted: Discovery × 0.30 + Conformance × 0.40 + Capabilities × 0.30. Discovery rewards HTTPS, reachability, agent-friendly robots.txt, and surface signals (llms.txt, sitemap, OG tags, Organization JSON-LD, viewport). Conformance rewards profile validity (3× weighted), spec version, and signing keys. Capabilities rewards declared transports, checkout, payment handlers, and (for opt-in checks) live endpoint responses.

What is a good UCP Score?

A (85+) means agents can reliably find, parse, and transact with your store — a high bar, deliberately calibrated to match Lighthouse and SSL Labs. B (70–84) is solid — minor gaps, agents will still work. C (50–69) is partial — profile works but missing capabilities or surface signals. D (30–49) is weak — profile reachable but invalid or near-empty. F (0–29) is failing — blocked, unreachable, or no profile detected.

Is checking my UCP Score free?

Yes. UCP Checker is free for any domain. No signup, no credit card. You can optionally save reports for monitoring (free email alerts when your score changes), but the score itself is always free and publicly indexed.

What does UCP_MISSING_LLMS_TXT mean?

This warning means no /llms.txt file is published at the root of your domain. llms.txt is an emerging convention (analogous to robots.txt) where you publish a hand-written brief aimed at LLM agents — about your brand, product categories, return policy, and any capabilities or restrictions. It's a low-effort discovery signal: missing it costs a small amount of Discovery score.

UCP Score

What's your UCP Score?

Lighthouse for agentic commerce. A 0–100 grade measuring how agent-ready your store is — used by AI shopping agents to decide which stores to surface, recommend, and transact with.

23,875

domains scored

79/100

average score

18,837

grade A stores

Grade distribution

Across 23,875 verified stores with a published, evaluable profile.

A 85–100 · Ready for agents

18,837 (78.9%)

B 70–84 · Nearly there

730 (3.1%)

C 50–69 · Partly ready

49 (0.2%)

D 30–49 · Hard for agents

522 (2.2%)

F 0–29 · Not ready

3,737 (15.7%)

Top 10 by score

The most agent-ready stores right now. See full leaderboard →

How it works

Three categories, one score

Every store is rated 0–100 in three areas, then combined into a single composite. Validity dominates — a broken profile can't be saved by surface polish.

30%

Agent Discovery

Can agents find and reach this store? HTTPS on the profile endpoint, reachability, agent-friendly robots.txt, plus surface signals: /llms.txt, sitemap, Open Graph, Organization JSON-LD, mobile viewport.

40%

UCP Conformance

Does the profile validate against the UCP spec? Validity is weighted 3× — an invalid profile cannot score above ~50 here. Also rewards declaring a known spec version, providing signing keys, and minimising warnings.

30%

Capability Coverage

What can an agent actually do? Rewards declared transports (REST / MCP / A2A), checkout, payment handlers, and breadth of capabilities. Functional probes confirm declared endpoints actually respond.

Read the full methodology →

Monitoring

Watch your UCP Score over time

Save any score report and we re-run the full check weekly. Email you only when something material changes — score drops, capability regresses, status flips. Free, no marketing.

assistant-experiments.vercel.app

A

100/ 100

100–100 · 30d

Last checked 14 hours ago

Last checked 1 day ago

Last checked 12 hours ago

Sign up to track your store's score over time. Same view, your domain.

How To

How to Get Your UCP Score

Free, no signup, no credit card. From domain entry to copy-paste fixes in under a minute.

1

Enter your domain

Type your store's domain (e.g. yourstore.com) into the form at the top of this page. We normalise the input automatically — protocols, www, paths, and trailing slashes are stripped before the check runs.

2

We crawl your profile & signals

We fetch /.well-known/ucp, validate it against the spec version your profile declares, capture the surface signals (llms.txt, sitemap.xml, Open Graph, JSON-LD, viewport), and run lightweight probes against each declared transport. All read-only.

3

Get your 0–100 score

Within seconds you get a composite UCP Score with category breakdowns for Discovery, Conformance, and Capability Coverage, plus an A/B/C/D/F grade and a shareable /score/{domain} URL.

4

Apply copy-paste fixes

Every gap on your /score/{domain} page comes with a ranked recommendation including impact, effort, and a copy-paste snippet. Drop them into your profile, robots.txt, sitemap, or HTML <head>, redeploy, and re-run.

What we score

Every signal we evaluate

The UCP Score combines static analysis, public surface signals, and (for opt-in checks) lightweight functional probes. Each signal feeds one of the three weighted categories.

Agent Discovery 30%

HTTPS on the profile endpoint
Reachability — profile returns 200 OK
robots.txt allows GPTBot, Google-Extended, ClaudeBot, Applebot-Extended, CCBot
/llms.txt brief for LLM agents
sitemap.xml at the root of the domain
Open Graph meta tags on the homepage
Organization JSON-LD on the homepage
Mobile viewport meta tag

UCP Conformance 40%

Profile validates against the declared spec version (3× weighted)
Spec version is a known, published release
signing_keys present and at the correct location for the version
No fatal errors (UCP_MISSING_VERSION, UCP_MISSING_SERVICES, etc.)
Minimal warnings (unused fields, unknown extensions)
HTTPS on all endpoint URLs declared in the profile

Capability Coverage 30%

Declared transports: REST, MCP, A2A, embedded
Checkout capability declared
Payment handlers declared
Breadth of declared capabilities
Probes (opt-in): transport endpoints respond to a benign request
Probes (opt-in): every spec and schema URL resolves

What we don't score

Page speed beyond profile fetch latency
SEO ranking signals (we score agent-readiness, not Google ranking)
Product catalogue completeness (that's UCP Playground's job)
/llms.txt body content — captured for monitoring, not scored
Brand reputation, reviews, or trust signals

Full ruleset documented in our methodology. For paste-and-validate without crawling, use the UCP Validator. To audit what your catalog endpoints return, use the Catalog Validator.

Live distribution

UCP Spec versions across scored stores

Distribution of declared spec versions across the 23,875 domains we currently score. Older versions are valid but lose points in the Conformance sub-score versus the latest release.

v2026-08-2511,204 stores v2026-04-088,355 stores v2026-01-1124 stores v2026-01-2316 stores

v1.06 stores

v2026-08-012 stores

v2025-01-011 store

v2026-05-291 store

v0.11 store

v1.0-VITL1 store

v2026-04-171 store

v1.0.01 store

Issue codes

Common UCP Score issues and how to fix them

The most common error and warning codes we surface across 23,875+ scored domains. Each code is a deep link from your /score/{domain} report to the specific copy-paste fix.

High-impact issues (drag your score below B)

Code	What it means	How to fix
`UCP_BLOCKED_BY_FIREWALL`	Your profile endpoint returns 403 — agents are firewalled out	Whitelist `/.well-known/ucp` or allow unauthenticated GET requests through your WAF.
`UCP_BLOCKED_BY_ROBOTS`	`robots.txt` disallows AI agents from your profile	Add explicit `Allow: /.well-known/ucp` for GPTBot, Google-Extended, ClaudeBot, Applebot-Extended, and CCBot.
`UCP_INVALID_MANIFEST`	Profile exists but fails schema validation	Run the UCP Validator against your JSON to surface specific field errors. Validity is 3× weighted in Conformance.
`UCP_PROBE_TRANSPORT_FAILED`	One or more declared transport endpoints don't respond when probed	Verify each declared `endpoint` resolves, returns 200, and (for MCP) accepts a `tools/list` request.
`UCP_INSECURE_ENDPOINT`	One or more endpoints are declared as `http://` (not HTTPS)	Change all `endpoint` URLs to `https://`. Agents reject non-HTTPS endpoints outright.

Surface signal warnings (cost ~3–5 points each)

Code	What it means	How to fix
`UCP_MISSING_LLMS_TXT`	No `/llms.txt` brief for LLM agents	Publish a plain-text brand brief at `/llms.txt`. Sections we look for: About, Product Categories, Policies, Agent Capabilities.
`UCP_MISSING_SITEMAP`	No `sitemap.xml` at the root of the domain	Most ecommerce platforms generate this automatically. If yours doesn't, generate one server-side and serve at `/sitemap.xml`.
`UCP_MISSING_OG_TAGS`	Open Graph meta tags missing or incomplete on the homepage	Add `og:title`, `og:type`, `og:url`, `og:description`, `og:image`, and `og:site_name` inside `<head>`.
`UCP_MISSING_ORG_SCHEMA`	No `Organization` (or `OnlineStore`) JSON-LD on the homepage	Add a JSON-LD block declaring your brand identity — name, URL, logo, sameAs (social profiles), contactPoint.
`UCP_MISSING_VIEWPORT`	No mobile viewport `meta` tag	One line inside `<head>`: `<meta name="viewport" content="width=device-width, initial-scale=1">`.
`UCP_VERSION_OUTDATED`	Profile declares a valid but older spec version	Upgrade to the latest published spec version — typically a small change but unlocks newer capabilities.

Full code list documented in our methodology. For an actionable fix recipe per code, run a check on any domain — the /score/{domain} report includes copy-paste snippets for every gap.

Who reads these signals

AI agents that consume UCP signals

These are the agents currently reading the surface signals, profiles, and capability declarations the UCP Score evaluates. A high score is a precondition for being recommended, surfaced, or transacted with by any of them.

Google AI Mode

Google's agentic commerce surface uses UCP profiles and Organization schema to enable shopping directly from search results. Stores without a valid profile are invisible to AI Mode shopping. Crawler: Google-Extended.

ChatGPT (OpenAI)

OpenAI's shopping agents use UCP for product discovery and checkout. GPTBot crawls /.well-known/ucp and /llms.txt to index your store's capabilities and brand brief.

Claude (Anthropic)

ClaudeBot reads UCP profiles when Claude is asked shopping or research questions about your brand. Claude is a heavy consumer of /llms.txt for context grounding.

Apple Intelligence

Applebot-Extended indexes UCP signals for Apple's on-device agent. Mobile-originating sessions render pages in a sandbox — the viewport meta tag matters here.

Common Crawl

CCBot archives profiles and surface signals; many downstream LLMs and agent products are trained or grounded on its corpus, so allowing CCBot indirectly improves visibility everywhere.

Shopify Shop app

Shopify's native agent reads UCP profiles for cart, checkout, and product data, with deep integration into the Shop app and Shopify Magic surfaces.

Platform comparison

Average UCP Score by ecommerce platform

Platform performance dominates outcomes — the median Shopify store has a very different infrastructure profile from the median WooCommerce store. Compare your score against your platform peers.

See full platform breakdown →

FAQ

Common questions

What is a UCP Score?

A 0–100 grade measuring how agent-ready a store is, computed from its UCP profile, surface signals, and lightweight functional probes. Scores combine three weighted categories — Agent Discovery (30%), UCP Conformance (40%), Capability Coverage (30%) — and map to A/B/C/D/F grades.

What's a good UCP Score?

A (85+) means agents can reliably find, parse, and transact with your store — calibrated to match Lighthouse and SSL Labs, so the A grade is a real signal. B (70–84) is solid — minor gaps. C (50–69) is partial — profile works but missing capabilities or surface signals. D (30–49) is weak — profile reachable but invalid or near-empty. F (0–29) is failing — blocked, unreachable, or no profile detected.

How is the score calculated?

Each category is rated 0–100 then weighted: Discovery × 0.30 + Conformance × 0.40 + Capabilities × 0.30. Validity is 3× weighted in Conformance — an invalid profile cannot score above ~50 in that category, regardless of how good the surface signals are. Full methodology →

Is checking my score free?

Yes. Every check is free, no signup required. You can save reports against your account if you want to track score changes over time, but the score itself is always free and publicly indexed.

How often is my score refreshed?

Domains in the directory are re-checked at least every 24 hours. Click "Run a fresh check" on your score page to force a real-time update with functional probes. Saved reports are also re-run weekly automatically.

Will my domain appear publicly?

Yes — every checked domain gets an indexed score page at /score/{domain}. This is intentional: public scores create transparency and a baseline for the agentic commerce ecosystem, in the same way SSL Labs grades and PageSpeed scores work.

How do I improve my UCP Score?

The score page lists prioritised, copy-paste fixes ranked by impact and effort. Most stores can lift their score by 10–20 points in under an hour by adding the surface signals (llms.txt, sitemap, Open Graph, Organization schema) and resolving profile warnings. Bigger lifts come from declaring more capabilities, adding transport endpoints (MCP, REST, A2A), and ensuring all declared endpoints actually respond when probed.

Which AI agents read these signals?

Google AI Mode, ChatGPT and OpenAI's GPTBot, Anthropic's Claude (ClaudeBot), Apple Intelligence (Applebot-Extended), Common Crawl (CCBot), Shopify's Shop app, and any agent built on the Model Context Protocol (MCP) or REST transport. The score reflects what these agents actually look for when deciding whether to recommend or transact with your store.

What does UCP_BLOCKED_BY_FIREWALL mean?

Your profile endpoint is returning a 403 — your WAF or origin is blocking unauthenticated GET requests to /.well-known/ucp. Whitelist that path or allow unauthenticated GETs through. Until you do, the score for the affected store is locked because we couldn't actually evaluate the profile.

How is the UCP Score different from UCP Status?

UCP Status is the binary verdict — does the profile exist, validate, and respond. The Score is a 0–100 grade that goes deeper into capability coverage, surface signals (llms.txt, sitemap, OG tags), and probe results. A store can be Verified-status and still have a low score if it only declares the minimum required capabilities. Most stores want both: green status, A-grade score.

How does this differ from the UCP Playground?

UCP Checker computes the score from your profile, surface signals, and lightweight probes — fast, safe, runs at scale. UCP Playground is a separate product that runs full end-to-end agent shopping simulations against your live infrastructure. Score tells you "can agents find and parse you" — Playground tells you "can agents complete a transaction".

Is the UCP Score the same as the Playground Score?

No. The Playground Score (Latency / Data Quality / Simulation Success) is a runtime score derived from live agent simulations in the UCP Playground. The UCP Score is computed by UCP Checker from static analysis plus lightweight probes. They're complementary; on this site, "Score" refers to the UCP Score unless explicitly noted.

Score your store now

Free, no signup. Enter any domain and get a 0–100 UCP Score with copy-paste fixes in under a minute.