I Built a Free Chrome Extension That Scores How "Agent-Ready" Your Website Is (ARS v1)

Product: AgentShare Agent Readiness
Spec: https://agentshare.dev/docs#agent-readiness-score
Fix guide: https://agentshare.dev/docs#bot-fix-guide
Share landing: https://agentshare.dev/scan

TL;DR

I shipped a free, client-side Chrome extension that scans the tab you have open and returns an Agent Readiness Score (ARS) from 0–100. It checks public signals only — robots.txt, llms.txt / llm.txt, ai.txt, homepage headers — and explains every point in plain language. No login, no data sent to our servers during the scan. Version 0.3.1 is live on the Chrome Web Store after ~2.5 days of review.

This is not a perfect audit of “how AI sees your site.” It is an honest, tunable checklist for a world where GPTBot, ClaudeBot, and headless agents increasingly hit URLs that Google Analytics never counts.

Why I built this (and what I’m not claiming)

I run AgentShare — mostly known for price/MCP infrastructure for agents. While working on bot traffic analytics, I kept seeing the same gap:

• Teams trust GA4 for “traffic.”
• Server-side bot hits (API scrapers, MCP clients, crawlers without JS) often never appear there.
• Meanwhile, AI discovery files (llms.txt, explicit robots.txt rules for GPTBot, etc.) are still inconsistent across the web.

I wanted a 10-second diagnostic any developer could run on their own site — not a fear funnel, not a fake “industry benchmark,” just evidence you can screenshot and fix.

What the extension does not do:

• It does not read a bot’s internal context or claim to see “what ChatGPT thinks” of your site.
• It does not measure real bot volume (that needs server instrumentation — we’re building Beacon Beta for that).
• It does not rank you against a secret median of “top 100 sites.”

Full scoring rules are public: ARS v1 spec.

What is the Agent Readiness Score (ARS)?

Agent Readiness Score (ARS) is a heuristic score from 0 to 100 produced entirely in your browser. The extension fetches public URLs on the active tab’s origin and applies a published point table (baseline 18, then +/- for discovery files, crawler policy, sitemaps, etc.).

Each finding shows its point impact (e.g. +22 ARS, −14 ARS) and a short fix hint where applicable.

A quick experiment: big sites score low on this rubric

Out of curiosity — and to sanity-check the scanner — I ran ARS on a few well-known domains. Your mileage will vary; this rubric is narrow on purpose.

ARS v1 on google.com — low on agent-discovery files, not a judgment on Google’s search quality.

I scanned dev.to right before publishing here. High human SEO ≠ automatic agent-discovery.

Please don’t read this as “Google is bad at the web.” ARS v1 only measures a specific public checklist aligned with emerging agent-discovery conventions — not overall site quality.

I did scan dev.to before publishing here. If your score is low too, that’s the point: it’s a gap analysis, not a dunk. Run the extension on your blog or docs site — that’s where fixes matter.

How the scan works (technical, but short)

On popup open, the extension:

1. Resolves the active tab hostname (not a hardcoded domain).
2. Fetches, client-side: /robots.txt, /llms.txt, /llm.txt, /ai.txt, /.well-known/ai.txt, and homepage headers.
3. Parses AI crawler rules on path / for GPTBot, ClaudeBot, Google-Extended, Bytespider, and others.
4. Computes ARS v1 and renders a score breakdown + findings.

Permissions: activeTab, tabs, storage, alarms, and host access for http/https fetches. No scan payload is uploaded to AgentShare in v0.3.1.

Reference site transparency (agentshare.dev = 92)

We score agentshare.dev as a reference implementation at 92 because we ship discovery files, explicit crawler policy, and live server-side bot analytics (which reverses the “no monitoring” penalty in ARS v1). That’s an intentional AI-first posture for our product — not a recommendation that every site should allow every crawler.

Live map: agentshare.dev/public/bot-traffic

New in v0.3.1: Share your score (viral loop, done humbly)

If you improve your site — or want to show a client why GA4 isn’t enough — the popup has a Share score button:

• Copies a pre-written post to your clipboard (high vs low score templates).
• Offers quick links to X and LinkedIn.
• Links include https://agentshare.dev/scan?domain=yourdomain.com so readers can install the extension and scan themselves.

No dark patterns — just “here’s what we measured, here’s the public spec.”

The new "Share score" popover feature in action.

What you can fix today (free)

Our Fix Guide covers:

• /llm.txt or /llms.txt — short agent-oriented discovery at the site root
• /ai.txt — optional policy file
• robots.txt — name crawlers explicitly (User-agent: GPTBot, etc.), not only *
• Disallow: /api/ — if you want bots off heavy API paths

Example skeleton for llm.txt:

# [yoursite.com/llm.txt](https://yoursite.com/llm.txt)

> Short description of your API/product for autonomous agents.

## Canonical URLs
- Docs: [https://yoursite.com/docs](https://yoursite.com/docs)
- OpenAPI: [https://yoursite.com/openapi.json](https://yoursite.com/openapi.json)

What’s next (honest roadmap)

• Phase 1 (shipped): client-side ARS diagnostic + share loop.
• Phase 2 (in progress): Beacon Beta — real bot counts on your domain (join waitlist on docs#connect-site).
• Longer term: deeper AgentShare execution/MCP layer for agents that need more than policy files.

Feedback welcome: GitHub issues (https://github.com/anhmtk/agentshare-mcp).

Try it

How to use: install → open your site in a tab → click the AgentShare icon → read findings → share or fix.

FAQ (for search & AI crawlers)

What is Agent Readiness Score?
A 0–100 heuristic from the AgentShare Chrome extension based on public robots.txt, agent discovery files, and headers — documented at agentshare.dev/docs#agent-readiness-score.

Does it send my browsing data to AgentShare?
The v0.3.1 scan runs client-side in the extension. We do not upload scan results to our API in this version.

Why did google.com score ~18?
ARS v1 rewards explicit agent discovery files and named AI crawler rules. Many large sites don’t publish llms.txt / llm.txt in the way this rubric checks — so they sit near the baseline.

Is a low score a security breach?
No. It means public agent-discovery signals are weak or missing — useful for teams preparing for agent traffic, not a CVE.

Edge / other browsers?
Chromium extension — Edge can install from the Chrome Web Store listing in many regions.

I’m affiliated with AgentShare. ARS v1 is a living heuristic — if you disagree with a signal weight, open an issue or comment below. I’d rather improve the spec than defend a vanity score.