“Make me a newspaper” — publishing a web page straight from a chat

What it actually looks like to ask an assistant for a web page, get a live URL on your own domain seconds later, then change your mind and have the same link update in place.

The slim.to team · · 4 min read

Here is the whole job, as it actually happened.

Create a sample link of a webpage that looks like a newspaper and publish it to slim.to with the slug of news. Use the domain share.kumkuat.ai

Twenty seconds later there was a URL. Not a file to download, not a preview pane, not a code block to copy — a live page on a real domain that anyone could open:

share.kumkuat.ai/news

A broadsheet-style newspaper page titled The Kumkuat Ledger, with a double-rule masthead, a dateline, a lead story with a drop cap, and three justified columns. Above it sits the slim.to viewer bar with the page title, an Identify yourself button, a Download button, and a Shared via slim.to badge.
The published page, on its own domain. A sample built to demonstrate the flow — the paper and its stories are fictional. The strip along the top is the slim.to viewer: the title, the optional identity prompt, a download button, and the badge you can switch off on a paid plan.

No dashboard was opened. No file was downloaded and re-uploaded. The assistant wrote the HTML and published it in the same breath, because slim.to was connected to it as a tool.

Then the inevitable happened

Looking at it, the obvious next thought:

Let's update the file to make it all news about Kumkuat

This is the moment that usually costs you. In the ordinary version of this story you regenerate the file, upload it again, get a new URL, and now you have two links — one of which you've already sent to somebody.

Instead:

Two turns. First: you ask for a newspaper page on your domain, the assistant calls create_link_from_content, and returns share.kumkuat.ai/news. Second: you ask for the content to be rewritten, the assistant calls replace_link_content, and the same URL is returned, marked unchanged.
Second draft, same address. replace_link_content swaps what the link serves. The URL, the domain, the access settings and the read history all survive.

Same link. New paper. Anyone already holding share.kumkuat.ai/news now sees the rewritten version, and the analytics didn't reset.

That's the part worth internalising, and it's the habit most people take longest to build. Publishing is cheap; re-publishing is what quietly ruins a shared link. One address that always has the current version beats five addresses and a guess about which one someone read.

What you're actually getting

It's easy to read this as "the assistant can make web pages", which undersells it. The page is a slim.to link, so it arrives with everything a link has:

The newspaper is a toy. The mechanism isn't. Swap "newspaper" for release notes, a status page, a client one-pager, a conference agenda, an internal FAQ — the flow is identical, and it's about four seconds of typing.

The bit that surprised me

One small thing from this run, worth knowing before it puzzles you.

replace_link_content swaps the file. It does not rename the link. So after the rewrite, the page's masthead said one thing while the link's display title in the dashboard still said what it was called on day one — which is also what shows in the viewer bar and in link previews.

Not a bug; it's the same reasoning that makes replace safe. The tool changes what it says on the tin, not the tin. If the title matters — and it does the moment someone pastes the link into Slack and an unfurl appears — update it in the link's config page, or ask for it explicitly:

"Also rename the link to The Kumkuat Ledger."

Setting this up yourself

Two minutes, one time:

  1. Add the slim.to connector to your assistant. On claude.ai that's Settings → Connectors → Add custom connector, then the URL https://slim.to/mcp, then Authorize.
  2. Add your domain in Dashboard → Domains and point a CNAME at us. TLS is automatic.

After that, "publish this at x" is a complete instruction. The full setup, including the desktop and CLI route for assistants that can read your local files, is in the MCP walkthrough.

Where to go next

Keep reading