How to publish a valid llms.txt on Replit

Verified 1 September 2026.

On Replit, publishing /llms.txt is not one job. It is two, and which one you have depends on how the app is published. An app built by Agent runs on an Autoscale Deployment with your own server answering every path, so the file is a build asset your server hands out. A Static Deployment has no server at all, so the file is a plain file in a public directory and you can set its content type from .replit. Getting this backwards is why the file looks published and then does not answer.

Either route takes about five minutes. Read the verification step twice, because the app scaffold Agent writes answers 200 for a file that is not there.

Quick summary

Find out which deployment type you have

Open .replit in the app root. According to Replit's configuration reference, these files "are hidden by default. Show them in your Replit App by selecting Show hidden files from the filetree menu." The value to read is deploymentTarget under [deployment].

An app scaffolded by Agent, fetched from the workspace on 1 September 2026, carries this:

[deployment]
deploymentTarget = "autoscale"
build = ["npm", "run", "build"]
run = ["npm", "run", "start"]

That is path A. If the Publishing tool shows a Deployment type of Static, that is path B. The two do not overlap, and Replit says so in its own words on the deployment types page: "Static Deployments are not compatible with Replit Apps created using Agent."

Path A: an Agent app on Autoscale

The React and Express scaffold Agent writes puts the front end under client/, and its vite.config.ts sets root to that folder with build.outDir pointing at dist/public. Vite copies the contents of the root's public/ folder into the build output untouched, so a file you drop in client/public/ arrives at the web root.

1. In the file tree, open client then public.
2. Create a file named llms.txt there. Not client/llms.txt, and not the repository root.
3. Write the file (the section below covers what belongs inside).
4. Click Publish, then Republish. The build step runs npm run build, which is what copies the file.

The server half is already done for you. The scaffold's server/vite.ts mounts express.static() on the build directory in production, so no route code is needed for a static file.

Path B: a Static Deployment

Here the public directory and an optional build command are set in the Publishing tool under Adjust settings. Put llms.txt at the top level of that directory, republish, and it answers at the root.

Static gives you one thing the other types do not: a content type you control. Add this to .replit in the app root:

[[deployment.responseHeaders]]
path = "/llms.txt"
name = "Content-Type"
value = "text/markdown; charset=utf-8"

According to Replit's static deployment guide, fifteen headers are reserved and cannot be set this way: Accept-Ranges, Age, Allow, Alt-Svc, Connection, Content-Encoding, Content-Length, Content-Range, Date, Location, Server, Set-Cookie, Trailer, Transfer-Encoding and Upgrade. Content-Type is not among them, so this rule takes effect after you republish. Most platforms in this series cannot do that at all, and neither of Replit's own files uses it.

The 200 that means nothing

The scaffold's production server, read from its server/vite.ts and verified on 1 September 2026, ends with a catch-all that sends the app shell for anything it did not match:

app.use(express.static(distPath));
app.use("*", (_req, res) => {
  res.sendFile(path.resolve(distPath, "index.html"));
});

A request for /llms.txt on an app with no such file therefore returns HTTP 200 with text/html and your homepage in the body. A checker that reads the status code alone will report success. So will one that follows redirects and stops there, which is the same class of false pass that Ghost produces by a different mechanism, covered in the Ghost guide.

Check the content type and the first line, never the status code by itself.

What Replit's own two files show

The format is small: an H1 with the site name, an optional blockquote summary, then ## sections holding markdown links. Replit publishes two of these and they are worth reading, though not copying wholesale. Both fetched and verified on 1 September 2026:

Two things to take from that. The docs file carries forty ### headings, which is deeper nesting than the format asks for and a common way a generated file drifts, alongside the other mistakes covered in the llms.txt validator guide. And both point overwhelmingly at .md targets rather than HTML pages, which is the part most hand-written files miss. If you would rather not draft one, the llms.txt generator writes a spec-shaped file you can paste in.

Confirm it with a scan, then by hand

Run the scan against the published domain:

npx aiscan-cli yourapp.replit.app

Read row C2, llms.txt, in the content dimension. It reports whether the file exists, returns 200, and contains an H1, at least one ## section and markdown links. AIScan grades it free and without an account, and the same pass covers C1, C3 and E3 on the same page.

If you would rather check by hand, ask for headers rather than a status code:

curl -sI https://yourapp.replit.app/llms.txt | head -3
curl -s https://yourapp.replit.app/llms.txt | head -1

The first should not say text/html. The second should be your # heading, not <!DOCTYPE html>. On the workspace preview URL the same commands work, but grade the published domain, since that is what agents fetch.

Where AIScan fits, and where it doesn't

A generated file that lists routes you deleted still passes C2. Re-read it when the app changes.

Publish, then read row C2

Add the file, republish, and scan the published domain. If C2 still fails, the file is in the repository root instead of client/public, or the build did not run. The same scan gives you C1, C3 and E3 on the same page, and the Astro guide covers the build-time version of this same job if you also ship a static site. Every fix guide in this series lives at aiscan.site/guides, the check itself is documented under content checks, and the Replit platform page collects what else we grade on this stack.