Skip to content

For agents /llms.txt

The agent-readiness score

Agents are reading your site.Can they use it?

Like Lighthouse, for AI agents. Themelior reads a page the way GPTBot and ClaudeBot do, plus its robots.txt, sitemap and llms.txt, and gives you a score out of 100 for whether agents can find it, act on it and get in safely. A free account shows the fix for each finding, and during the beta the change written for that page.

  • About ten seconds
  • Real requests, real checks
  • No account needed
  • Report public by its link, never indexed

Try

Agent-readiness report: themelior.com

Scanned 13 Nov 2026 · 21:55 UTC

100 out of 100Static score

  • Find100
  • Act100
  • FinishNot measured. Runs when the verified site owner asks
  • Safe100
  • Stays correctNot measured. Needs a second scan

Nothing to fix in the static checks

Every static check passes: Find, Act and Safe. Finish needs real agents and Stays correct a second scan, so neither is measured here.

Static score 100 out of 100.

Open the report

The YC sample · 24 Oct 2026

We scanned 100 new startups.Three passed every static check.

fail at least one static check. Static checks only: no real agent was sent to these pages.
97/100
give agents no machine interface: no MCP server, no OpenAPI, nothing linked.
94%
have no llms.txt, the plain-text summary an agent reads first.
47%

Home pages of 100 Y Combinator companies from the latest batches, scanned with the same engine as your report and scored by method 1. Three served a bot challenge instead of their home page, so 94% and 47% are shares of the 97 readable home pages. Aggregates only, so no company is named. The sample, the method and the limits.

How it works

01 / Scan

Read like an agent.

We send the requests an agent sends: the page, robots.txt, the sitemap, llms.txt, the MCP and OpenAPI descriptions. Five checks, fourteen sub-checks, one score out of 100. Every lost point says what we saw.

Stream · NDJSONexample.com · 31 Oct 2026 · method 1.1
{"type":"step","step":"page","found":true,"detail":"200 · 1 KB"}{"type":"step","step":"robots","found":false}{"type":"step","step":"llms","found":false}…four more steps{"type":"step","step":"score","found":null,"detail":"45"}{"type":"done","id":"a726f494…","score":45,"reused":false}
Find
Can an agent discover the page and what it is for.
Act
Can it read the structure and press the controls.
Safe
Can it get in without being treated as an attack.
Finish
Can a real agent finish a real task. Measured, not guessed.

02 / Fix

The change, written for your site.

Not advice. The file, the tag, the line, ordered by the points each one adds. Copy it in, or send it to your repository as a pull request once it is connected. Free with an account during the beta.

/robots.txtillustrative
User-agent: *Disallow: /cart- User-agent: GPTBot- Disallow: /+ User-agent: GPTBot+ User-agent: ClaudeBot+ Allow: /Sitemap: https://shop.example/sitemap.xml
Ordered
By the points each fix adds.
Exact
Written against what we read on your page.
Shippable
Copy it, or open a pull request.
Checked
“Check my fixes” proves it landed.

03 / Finish

Real agents. Real tasks.

A static check can only guess. Finish sends real agents to answer real questions from your own pages, and records where each one got stuck and why. Runs on your own Anthropic model key.

Finish traceillustrative
00:00 open https://shop.example/00:02 find “search” → input[name=q] ✓00:04 read /catalogue/trail-runner-200:06 answer “Which sizes ship today?”00:06 the sizes are drawn by a script✗ stuck: no sizes in the HTML served→ fix: render the size list on the server
Measured
The full score counts what agents finished.
Diagnosed
Every failure names the step and the cause.
On demand
When you ask, once a day, on the site you proved is yours.
Your key
Runs on your own Anthropic model key.

04 / Agents · closed beta

Then serve them yourself.

Once agents can use your site, give them a way in that you control: a site agent, built from your pages, served over MCP and answering with sources. It is in a closed beta: today only the people who run Themelior build one, on their own site. Verified site owners come next, with no date yet.

Want one for your site? Join the waitlist: it counts who asks and charges nothing; we write only when it can be used.

Live demo · one callpractice site · 31 Oct 2026
→ books_toscrape__find_books{"category":"Mystery","minRating":5,"sortBy":"price-asc","limit":5}← 5 matched · read index.html, page-2.htmlThe Girl You Lost · £12.29The Silkworm (Cormoran Strike #2) · £23.05What Happened on Beale Street (Secrets of the South Mysteries #2) · £25.37…two more books
Yours
Built from your pages, nothing else.
Sourced
Every answer links where it came from.
Standard
MCP, so any agent can call it.

05What a scan reads

Fourteen sub‑checks. Every point accounted for.

Each card is one sub-check and the points it carries in its check of 100. Swipe or scroll the rail sideways (Shift and the wheel), or tab to it and use the arrow keys.

  1. Find01 / 14

    AI crawlers

    robots.txt lets GPTBot, ClaudeBot and the rest read the page.

    30points of Find’s 100

  2. Find02 / 14

    Sitemap

    A sitemap lists the public pages, named in robots.txt or at /sitemap.xml.

    20points of Find’s 100

  3. Find03 / 14

    llms.txt

    A plain-text summary an agent reads first.

    20points of Find’s 100

  4. Find04 / 14

    Structured data

    JSON-LD says what the page is in schema.org terms.

    20points of Find’s 100

  5. Find05 / 14

    Title and description

    A real title and meta description.

    10points of Find’s 100

  6. Act06 / 14

    Named controls

    Every link, button and field has a name an agent can read.

    35points of Act’s 100

  7. Act07 / 14

    Headings

    One h1, the rest nested without skipping.

    15points of Act’s 100

  8. Act08 / 14

    Landmarks

    A <main> landmark around the content.

    15points of Act’s 100

  9. Act09 / 14

    Machine interface

    An MCP or OpenAPI description, linked or at a known path.

    15points of Act’s 100

  10. Act10 / 14

    Language

    The page declares its language.

    10points of Act’s 100

  11. Act11 / 14

    Image text

    Every image carries alt text.

    10points of Act’s 100

  12. Safe12 / 14

    Open to agents

    No wall in front of a public read.

    40points of Safe’s 100

  13. Safe13 / 14

    HTTPS

    Served over HTTPS, with a redirect from HTTP.

    30points of Safe’s 100

  14. Safe14 / 14

    No CAPTCHA on read

    Reading a public page needs no CAPTCHA.

    30points of Safe’s 100

06Our own site

We ran it on ourselves first.

themelior.com scored 67 on its first scan. Six fixes to the site took it to 90: 23 points. Then our own CAPTCHA check matched our page’s text about CAPTCHAs; we corrected the check, and the last 10 points followed: 100.

Static score, method 1.1, last read on 9 Nov 2026. Other tools score this page differently.

  1. 67 out of 100

    First scan
  2. 90 out of 100

    Six fixes
  3. 100 out of 100

    Check corrected

07Questions

Short questions.Straight answers.

01Is it free?

A scan of any public page is free with no account. A free account reads the full report with every fix, and during the beta it also keeps your sites watched. Paid plans have no price yet, and the beta no end date. Before anything you use today costs money, we email you at least 30 days ahead. Free after the beta, too: a scan of any public site with no account, and its report preview; your account and its reports, with their history: the beta’s end deletes nothing stored; fix pull requests already opened; exporting or deleting your data; asking us to remove a report about your site. More on Pricing.

02What exactly does a scan request?

Eight requests, every one a GET, as ThemeliorBot/1.0: your page, /robots.txt, /sitemap.xml, /llms.txt, /.well-known/mcp.json, two OpenAPI paths and your page again over plain HTTP, to see that it moves to HTTPS (seven requests when your page is plain HTTP: that check is skipped). Nothing is submitted or changed. The list is on ThemeliorBot.

03Who can see a report?

Anyone with its link. Every scan makes a report at an address no one can guess; it is kept out of search engines and stays up. The verified site owner can unlist the site’s reports, or ask us to remove one through Contact.

04Static score or full score?

The static score is Find, Act and Safe, read from your files. The full score adds Finish and Stays correct, which need real agents and history. Every report says which one you are reading.

05Can I keep ThemeliorBot out?

Yes: block its user agent at your server or CDN and every request stops at once. robots.txt does not stop a scan someone asked for; it is one of the files the scan reads. To have scans of your site stopped and its reports unlisted, write to us and show the site is yours: ThemeliorBot says how. Once you verify your site in a free account, you can unlist its reports there yourself.

06Do you change my site?

Never on our own. Fixes are text you copy or a pull request you review and merge.

08Your turn

Ten seconds.Then you’ll know.

Type a domain. The requests we will send appear beside it.

Requestsyour-site.com
  1. GEThttps://your-site.com/
  2. GEThttps://your-site.com/robots.txt
  3. GEThttps://your-site.com/sitemap.xml
  4. GEThttps://your-site.com/llms.txt
  5. GEThttps://your-site.com/.well-known/mcp.json
  6. GEThttps://your-site.com/openapi.json
  7. GEThttps://your-site.com/.well-known/openapi.json
  8. GEThttp://your-site.com/ → https?