On October 26, Anthropic, the company behind Claude, filed to list on the Nasdaq under the ticker "ANT." (CNBC) Whatever you think of the valuation, the message for SaaS teams is simple: AI assistants and the agents built on them aren't a side experiment anymore. They're turning into a front door to software, the way search engines did twenty years ago.
And just like search, an agent can only use what it can find and understand. If an agent can't figure out how to work with your product, it picks one it can. Your customer never files a ticket or cancels. They just never show up. We call that ghost churn.
So how does an agent decide whether your product is usable? It doesn't read your marketing page. It looks for machine-readable signals.
The 7 signals
We built a free scanner that checks a domain for seven public signals. Here's what each one is and why it matters.
/llms.txt: a plain-text guide at the root of your site that tells language models what your product is and where the important docs live. It's the agent equivalent of a sitemap./llms-full.txt: the expanded version, with full documentation content in one file so an agent can load it without crawling dozens of pages.- AI agents allowed in
robots.txt: if your robots rules block AI user agents, assistants can't read your site at all, no matter how good your docs are. - An OpenAPI or Swagger spec: a machine-readable description of your API, published at a standard path, so an agent knows exactly which endpoints exist and what they take.
- Public API docs: human-readable documentation agents can also read, so they understand concepts, auth and limits.
- A published MCP server: the Model Context Protocol is the open standard assistants like Claude, ChatGPT and Cursor use to connect to tools. A discoverable MCP server means an agent can actually act in your product, not just read about it. GitLab, Traefik and Interact all shipped ones recently — we covered that wave in GitLab, Traefik and Interact shipped MCP servers.
- schema.org structured data: JSON-LD markup on your homepage that tells machines who you are and what you offer.
The two most heavily weighted signals are whether your robots rules let AI agents in and whether you publish an OpenAPI spec, because without them an agent can't read your site or call your API reliably.
How well-known products score
We ran the scanner on a few domains, ourselves included.
| Domain | Score | What the scan found |
|---|---|---|
| stripe.com | 68 | llms.txt, AI agents allowed in robots.txt, public docs, structured data; MCP mentioned in docs but no server found at the standard paths we check |
| gitlab.com | 68 | llms.txt, AI agents allowed in robots.txt, public docs, structured data; MCP mentioned but no server found at the standard paths we check |
| pryno.ai | 53 | llms.txt, AI agents allowed in robots.txt, structured data; no public API docs or OpenAPI spec |
Two things stand out.
First, even companies with excellent developer platforms don't max out. Both Stripe and GitLab run public MCP servers. Stripe documents its own at mcp.stripe.com (Stripe), and GitLab made its server generally available on October 15 (GitLab). Our scan didn't find either one at the standard discovery paths on their main domains. Building an MCP server is one step. Making sure agents can discover it is another.
Second, yes, we scored 53 on our own scanner. We're a services company without a public API, so the API signals don't apply to us the way they do to a SaaS product. We'd rather show you an honest number than a flattering one.
One caveat: the scanner only reports what it can detect at the URLs it checks. A missing signal means we didn't find it there, not that it can't exist somewhere else.
Check your own product
Enter your domain, and you'll get your score and a list of what was and wasn't detected right away. Leave your email to unlock the full report, with the evidence URLs we checked and specific recommendations.
Run the free Agent Readiness Check →
If your score is lower than you'd like, that's what Pryno does. We build MCP servers and AI integrations so products can be used by agents like ChatGPT, Claude and Cursor. Think of it as SEO, but for AI.
