What is an Agent Readiness Score?

A single 0-100 grade covering three things: whether an assistant can get in at all, whether what it finds is valid against the UCP and WebMCP specs, and how much of the site it can actually act on. One number, with the specific gaps listed underneath.

Does Orbilo check both UCP and WebMCP?

Yes, in the same check. UCP is the commerce manifest at a fixed address on the domain; WebMCP is the set of tools a page offers an agent in the browser. A site can pass one and fail the other, so both count toward the score.

Do I need a developer to run this?

Not to run the check, which needs only your web address. The fixes are usually settings rather than new code, and the report explains each one in a form you can forward to whoever looks after your site.

Is this just SEO with a new name?

No. SEO decides whether a person finds a site in a list of links. This decides whether software can understand what a site offers and then act on it. A site can rank first and still be unusable to an assistant.

Is it too early to worry about this?

UCP launched in January 2026 with Google, Shopify, Etsy, Wayfair, Target and Walmart. WebMCP is a W3C proposal shipping in Chrome. Assistant-driven transactions are already running on a major search platform, and these files are being published today. Checking where a site stands costs a minute.

Checking is free and needs no account. Monitoring one site is free permanently, not as a trial. Paid plans start at $29 a month for teams watching several sites.

How do I know the check is right?

The checks are open source. You or your developer can read exactly what they test and run them from a terminal. The hosted version uses the same scoring, so the two cannot disagree.

What happens after the first check?

Nothing, unless you ask for it. Add a site to monitoring and Orbilo re-checks it daily and tells you when the score moves, when a spec version changes, or when something that was working stops.

NEW WebMCP checks are live

People are sending AI assistants such as ChatGPT, Claude, Google, Perplexity, Copilot instead of visiting your site People are sending ChatGPT Claude Google Perplexity Copilot
instead of visiting your site

Can AI agents understand your site? If they cannot, they leave, and your competitor gets the traffic.

Free · no account · nothing stored unless you ask

The web just got a third kind of visitor

For thirty years every visit came from one of two things: a person, or a crawler indexing pages so a person could find them later. Sites are built for the first and tolerant of the second.

Agents are the third kind, and they fit neither mould. They arrive like a person, on someone's behalf with a task to finish, but they read like software. They do not scroll around a missing price, guess what a button does, or come back later to try again.

A site can work beautifully for people and crawlers and still be unusable to agents. When that happens, the agent finishes the task somewhere else, and nothing in your analytics records that it ever visited.

SCORE Agent Readiness Score

Know exactly what needs fixing.

One 0–100 grade that tells you where your site stands with AI agents: three weighted categories rolled into one number, and every gap comes with its fix. Monitor your site and the score is tracked over time, so a change that breaks something shows up as a number moving, not as a customer complaining.

Run a check

your-site.com

Agent ready

84 /100

Access 30%: 92
Conformance 40%: 88
Capability 30%: 71

Know what agents see. Know when it changes.

Your site looks fine to you because you already know what it does. An agent arrives knowing nothing, and the gap between the two is invisible from the inside. The check makes it visible; monitoring keeps it that way.

See what agents see

Everything your site declares to an agent, fetched the way an agent fetches it and validated field by field, with a fix for every gap.

Know the moment it changes

Your site is re-checked every 24 hours with full history, and you get an alert the moment it breaks, recovers, or changes spec version.

A moving target, handled

These standards are months old and still changing. We track every release so you do not have to, and re-check your site against whatever version agents are using now.

What agents need from your site

Three things decide whether an agent can use your site. Most sites need two of them, and the check tells you which.

UCP

/.well-known/ucp

The Universal Commerce Protocol: one manifest that tells shopping agents what you sell and how to buy it. Without it, a store does not exist to them. It is commerce-specific. What this means by sector.

What is UCP? →

WebMCP

navigator.modelContext

The Web Model Context Protocol: real buttons for agents instead of guesswork. Your page hands them search, filter and book directly, and it works on almost any interactive site.

What is WebMCP? →

Access

robots.txt · rate limits · WAF

Bot rules and rate limits turn agents away by default, and it looks exactly like security working. The most common failure of the three, and usually a one-line fix.

Which agents to allow →

Everything you need to get agent-ready

From the first check to daily monitoring: one place for your whole agent-facing surface. No SDK, no code changes.

Check

Protocol Detection: Point Orbilo at any domain. We locate the manifest and the page tools and report exactly what agents can see.
Spec Validation: Field-level validation against the live UCP and WebMCP schemas, with a copy-paste fix for every error.
Access Audit: robots.txt, rate limits and bot rules across GPTBot, ClaudeBot, Google-Extended and the rest.

Monitor

Daily Re-checks: Every monitored site re-checked every 24 hours, with full status history.
Change Alerts: Notified the moment a site breaks, recovers, or changes spec version.
Spec Changelog: Every UCP and WebMCP release tracked as it lands.

Understand

Sector Guides: What agent readiness means for shops, bookings, SaaS and services, sector by sector.
Agent Directory: Every major agent, how it identifies itself, and how it behaves.
Free Tools: llms.txt and JSON-LD generators, no account needed.

Being skipped is silent

When an agent cannot work out what a site offers, it uses another one. No bounce, no failed search, no line in any report: a site that was skipped looks exactly like a site nobody was looking for.

Why this happens, in full

Questions

What is an Agent Readiness Score?: A single 0-100 grade covering three things: whether an assistant can get in at all, whether what it finds is valid against the UCP and WebMCP specs, and how much of the site it can actually act on. One number, with the specific gaps listed underneath.
Does Orbilo check both UCP and WebMCP?: Yes, in the same check. UCP is the commerce manifest at a fixed address on the domain; WebMCP is the set of tools a page offers an agent in the browser. A site can pass one and fail the other, so both count toward the score.
Do I need a developer to run this?: Not to run the check, which needs only your web address. The fixes are usually settings rather than new code, and the report explains each one in a form you can forward to whoever looks after your site.
Is this just SEO with a new name?: No. SEO decides whether a person finds a site in a list of links. This decides whether software can understand what a site offers and then act on it. A site can rank first and still be unusable to an assistant.
Is it too early to worry about this?: UCP launched in January 2026 with Google, Shopify, Etsy, Wayfair, Target and Walmart. WebMCP is a W3C proposal shipping in Chrome. Assistant-driven transactions are already running on a major search platform, and these files are being published today. Checking where a site stands costs a minute.
What does it cost?: Checking is free and needs no account. Monitoring one site is free permanently, not as a trial. Paid plans start at $29 a month for teams watching several sites.
How do I know the check is right?: The checks are open source. You or your developer can read exactly what they test and run them from a terminal. The hosted version uses the same scoring, so the two cannot disagree.
What happens after the first check?: Nothing, unless you ask for it. Add a site to monitoring and Orbilo re-checks it daily and tells you when the score moves, when a spec version changes, or when something that was working stops.

Find out where you stand

One web address is all it needs, and the answer is plain either way.

Prefer to read first? What this means by sector, the guides, or which assistants are out there.

People are sending AI assistants such as ChatGPT, Claude, Google, Perplexity, Copilot instead of visiting your site People are sending ChatGPT Claude Google Perplexity Copilot instead of visiting your site