Agent-readiness scoring frameworkswhat each one actually measures.

Why five frameworks exist

Twelve months ago there was no scoring framework for agent-readiness at all. Today there are five. The reason isn’t that the problem fragmented — it’s that five different actors saw the same surface area and built scoring from different angles.

• Google (Chrome team) shipped Lighthouse Agentic Browsing because Lighthouse is where browser-side site-quality conversations already happen. Adding the category was the natural extension of Performance, Accessibility, Best Practices, and SEO.
• Cloudflare shipped isitagentready.com because Cloudflare sits in front of a large share of the public web and has a commercial interest in defining what “agent-ready” means at the edge.
• Google (Commerce / Shopping team) shipped Universal Commerce Protocol (UCP) at NRF in January 2026 with Shopify, Etsy, Wayfair, Target, and Walmart — a commerce-rail spec for AI agents transacting on merchant sites. It scores narrower than Lighthouse but is a Google-backed standard with serious retailer adoption.
• Bridge AI built an enterprise platform around the broader governance question (training-data permissions, content licensing, AI policy enforcement) rather than just file presence.
• The Optimisers, an established SEO/CRO agency, pivoted into the category with a bespoke Agentic Readiness Score they deliver inside a paid audit engagement.

The five are not competing for the same buyer. Lighthouse is for any developer running an audit. Cloudflare is for the Cloudflare-adjacent buyer. UCP is for commerce sites within the Google merchant orbit. Bridge AI is for enterprise. The Optimisers is for the agency-services buyer. Knowing which framework you’re looking at — and which one you’re actually being scored against — is the first useful piece of triage.

The five frameworks at a glance

Lighthouse Agentic Browsing — the open standard

Who runs it. Chrome team, Google. Shipped late 2025 as a fifth Lighthouse category alongside Performance, Accessibility, Best Practices, and SEO. Spec published at developer.chrome.com/docs/lighthouse/agentic-browsing/scoring.

What it scores. Nine specific audits today, grouped into four families:

• File presence — llms-txt-present, agents-json-present, agent-runbook-present
• File correctness — llms-txt-well-formed, agents-json-actions-typed
• Discovery — sitemap-discoverable, auto-discovery-links
• Content structure — schema-org-density
• Emerging — webmcp-annotations (low weight; spec still moving)

Where the weighting lives. Open and published. Each audit returns pass, fail, or a pass ratio rather than a single weighted 0–100 score. The audit list and scoring methodology are versioned on the developer.chrome.com docs.

Update cadence. Chrome ships audits to the category on the Lighthouse release cycle — typically every 4–6 weeks per Chrome version. The category is on track to leave experimental status in Q3–Q4 2026, at which point its audits become defaults in every Chromium browser’s DevTools.

Reach. Highest of any framework. Every Chromium browser (~71% of global browser sessions per StatCounter, April 2026) plus Edge (~5%). The audit category ships into the same DevTools tab every developer is already opening.

Cloudflare Agent Readiness Score

Who runs it. Cloudflare. Shipped at isitagentready.com around April 17, 2026.

What it scores. Four dimensions:

• Discoverability — robots.txt well-formedness,sitemap.xml reference, HTTP Link headers, crawler-enumeration hygiene
• Content — llms.txt presence, Markdown-for-agents fallback URLs, plain-text accessibility
• Bot Access Control — per-bot rules in robots.txt (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot), Content Signals, Web Bot Auth
• Capabilities — MCP Server Card at /.well-known/mcp/server-card.json, API Catalog discoverability, OAuth server discovery, Agent Skills declarations

Where the weighting lives. Cloudflare publishes per-dimension breakdowns alongside each scored URL, plus ships a coding-agent prompt for each failing check. The exact intra-dimension weighting is not externally documented in the same audit-level granularity as Lighthouse.

Update cadence. Cloudflare adds dimensions and refines existing ones on their own product release cycle. No public versioning is documented for the score itself.

Reach. Anyone who pastes a URL into isitagentready.com. The score has visibility inside the Cloudflare-engineer and SEO-practitioner communities but does not ship into a browser tab the way Lighthouse does. Strongest among buyers who have a coding agent (Claude Code, Cursor) and can act on the per-failure prompt.

The honest read. Cloudflare’s score is anchored heavily to its own product stack alongside open standards. MCP Server Cards and Agent Skills are real specs but Cloudflare-stewarded and not yet broadly adopted outside Cloudflare’s ecosystem. A small site that scores poorly on the Capabilities dimension may not reflect a meaningful gap in actual agent-readiness — it reflects a gap in Cloudflare-flavored agent-readiness.

Universal Commerce Protocol (UCP) — the commerce-rail framework

Who runs it. Google, with a launch consortium of Shopify, Etsy, Wayfair, Target, and Walmart. Unveiled at the National Retail Federation conference in January 2026. Spec is open at ucp.dev with implementation guidance at developers.google.com/merchant/ucp. Current version: 2026-04-08.

What it scores. UCP isn’t scored on a 0–100 scale like Lighthouse. It’s a binary: either your site exposes a valid manifest at /.well-known/ucp with declared capabilities and transport bindings, or it doesn’t. Agent surfaces that read UCP (Gemini, Google AI Mode, Google Shopping, the parts of ChatGPT browse that transact) discover what your site sells and how to transact through it. Agent surfaces that don’t read UCP fall back to general web crawl plus llms.txt / agents.json.

What the manifest declares.

• Organization metadata — name, URL, logo, description
• ucp_version — currently 2026-04-08, with 2026-01-23 still accepted
• Capabilities array — typed commerce actions: checkout, browse_products, inventory, payment-handler declarations
• Transport bindings — for each capability, which protocol it speaks: REST API, MCP (Model Context Protocol), or A2A (Agent-to-Agent). Multiple bindings allowed; agents pick the one they understand.

The Shopify shortcut. If you run a Shopify store, UCP is already shipped — Shopify auto-enables /.well-known/ucp on every store with an MCP transport binding at /api/ucp/mcp. Verify with:

curl https://yourstore.com/.well-known/ucp

The non-Shopify situation. WordPress / WooCommerce, Webflow, Squarespace, Wix, BigCommerce, Magento, and custom builds have no native UCP support as of May 2026. You can implement the manifest manually + host at least one transport binding, but this is engineering work measured in days, not minutes.

The honest read. UCP is real and Google-backed and you should care if you run a commerce site, but as a scoring framework it is narrower than Lighthouse — it audits one file (the UCP manifest) and one capability layer (commerce). For Shopify stores, UCP-readiness is auto-handled and the attention belongs on the four files UCP doesn’t cover (agents.json, llms.txt, agent-instructions.md, Schema.org density). For non-Shopify commerce sites, UCP becomes a manual project the kit doesn’t currently generate — wait for native platform support unless transaction volume justifies custom engineering. The deep-dive on how UCP, agents.json, MCP, and llms.txt complement each other lives at the three-protocols orientation map.

Bridge AI

Who runs it. Bridge AI (buildbridges.co). Pre-2026, currently active.

What it scores. A broader governance surface than the file-presence frameworks above. Bridge AI positions against AI usage policies, content-licensing controls, and agent-safe standards — closer to compliance tooling than to a developer audit. The platform’s specific scoring methodology is not publicly documented.

Where the weighting lives. Delivered as part of a paid engagement; not externally auditable.

Update cadence. Vendor-controlled.

Reach. Enterprise and mid-market buyers encountered via outbound sales motion. No self-serve path. Different market segment from the SMB self-serve buyer. Brand-name collision with BridgeToAgent is real; product collision is not — the segment cleanly separates.

The Optimisers — Agentic Readiness Score (ARS)

Who runs it. The Optimisers, an established SEO/CRO agency that pivoted into the agentic-readiness category in Q1–Q2 2026 (theoptimisers.com/agentic-readiness-audit/).

What it scores. A proprietary single-metric ARS plus a remediation roadmap. The metric’s component weighting and audit list are not publicly documented.

Where the weighting lives. Internal to the agency. Delivered as a deliverable inside the audit engagement.

Update cadence. Agency-controlled.

Reach. Agency clients buying an audit engagement. Different model from a $49 self-serve kit.

How the BridgeToAgent kit affects each score

Honest mapping. The kit ships three files (agents.json, llms.txt,agent-instructions.md) plus the auto-discovery <link rel="alternate"> snippet for your <head>. Here is what that fixes inside each framework.

Which one to act on first

If you’re an SMB or solo founder with a transactional site: Lighthouse first. It’s the audit that gates whether AI agents complete actions on your site, and the same handful of files that close the Lighthouse audits also contribute to Cloudflare’s Content and Capabilities dimensions.

If your concerns are governance-first — training-data permissions, AI policy enforcement, content licensing — Cloudflare or Bridge AI are the more direct fit, and the file-presence work is complementary rather than primary.

If you’re already inside an agency engagement that uses The Optimisers ARS: cross-reference the ARS components against the Lighthouse audits. Where they overlap, the kit moves both scores. Where they don’t, the agency roadmap will tell you what additional work is in scope.

Frequently asked

How often is this page updated?

Quarterly, or sooner if a framework ships material changes. Last reviewed 2026-05-21. We track each framework’s public-facing audit list and update the tables when a vendor adds or removes a dimension.

Is there a single “industry standard” framework?

Lighthouse Agentic Browsing is the closest thing. It’s stewarded by Google, ships into every Chromium browser, and anchors to open public specs that other vendors’ frameworks reference (llmstxt.org, the agents.json reference, Schema.org). When site-quality conversations happen at scale — the way Performance and Accessibility became default conversations — Lighthouse’s framework will be the reference point.

What is UCP and how does it compare to Lighthouse?

UCP (Universal Commerce Protocol) is a Google-backed open standard launched at NRF in January 2026 with Shopify, Etsy, Wayfair, Target, and Walmart. It declares a merchant’s commerce capabilities at /.well-known/ucp so AI agents can discover and transact. UCP is narrower than Lighthouse — it covers the commerce transaction layer only, not the broader file presence + content structure surface Lighthouse audits. Shopify auto-enables UCP on every store; non-Shopify platforms have no native support yet (as of May 2026). For most site owners, Lighthouse is the framework to optimize against today because it’s a default audit in every PageSpeed Insights run; UCP becomes relevant if you run a commerce site, in which case Shopify hands it to you for free. The longer comparison lives in the three-protocols orientation map.

Does the BridgeToAgent kit ship UCP support?

No, the kit does not currently generate /.well-known/ucp. For Shopify stores UCP is auto-enabled by the platform — the kit doesn’t need to add it. For non-Shopify platforms (WordPress, Webflow, Squarespace, Wix, custom), implementing UCP manually is engineering work in the days-not-minutes range, outside the kit’s scope. The kit instead generates the four files UCP doesn’t cover: agents.json (full action map beyond commerce), llms.txt (AI search context), agent-instructions.md (brand voice + escalation), and three <link rel="alternate"> tags for auto-discovery. The defensive Shopify view is at the UCP is Shopify-native explainer.

Why isn’t WebMCP Checker on this list?

WebMCP Checker (webmcp-checker.com) is a free single-purpose audit tool, not a scoring framework. It runs one specific test (WebMCP per-element annotation presence) and reports pass/fail. It maps onto the webmcp-annotations audit inside Lighthouse but isn’t a multi-dimension framework you compare against the four above.

Will more frameworks appear?

Likely yes through 2026 — every major platform (Vercel, Netlify, Shopify, WordPress) has a commercial incentive to define what “agent-ready on our platform” means. The differentiator over time will be whether a framework anchors to open public specs (durable) or to a vendor stack (vendor-aligned). We’ll add new frameworks to this page as they reach the audit-comparison threshold.

What if my framework score and my Lighthouse score disagree?

Common, and not a contradiction. Read the failing audits rather than the headline number. A site scoring 50 with the right kinds of audits passing is in better shape than a site scoring 70 with all the easy audits passing and the hard ones (agents-json-actions-typed, schema-org-density) failing. The pillar at Cloudflare vs Lighthouse — what each one actually measures walks through specific divergence scenarios on the same hypothetical site.