Search for "html to url" and the results answer three different questions. Sort out which one is yours, because two are quick and one is a different tool entirely.
- You have HTML and want a web address for it. This page.
- You have an address and want its HTML. That is view-source, or a converter like urltoany.
- You want to write a link inside HTML. That is the
<a href>tag, and MDN explains it better than anyone.
Everything below is the first one: code on your screen, a link you can send.
The prompt
If you already pay for Claude or ChatGPT, this is one message. Connect Birta to your agent once, then:
claude mcp add --transport http birta https://birta.dev/mcpThe endpoint is https://birta.dev/mcp, and the setup lives in the Claude Code guide. Then paste the prompt:
Take the HTML below (or the file at [path]), create a Birta project, write it in
as index.html, publish it, and give me the address.
Before publishing, tell me if it references anything that will not be there:
a stylesheet, a script, an image, a font. If it does, either inline it or tell
me what to hand you.
[paste your HTML here]The second paragraph saves the round trip. A single HTML file that references
style.css publishes fine and then looks broken, because the stylesheet never came with
it. An agent asked to look will tell you before the link exists.
What to check before you send the link
Open the address yourself first, in a browser you are not logged into anything with. Four things break most often:
- Missing sidecar files. CSS, JS, images and fonts that live beside the HTML have to be published too. One file in, one file out.
- Absolute paths.
/style.csswith a leading slash means "the root of this domain". Relative paths (style.css,./assets/logo.svg) survive being moved. - Case.
Logo.PNGandlogo.pngare the same file on macOS and two different files on a web server. This one bites quietly. - Local-only references.
file:///Users/you/...in ansrcworks on your machine and nowhere else.
If the code is in a chat, not a file
Same prompt, minus the file path: paste the code straight in. This is the common case with an artifact or a canvas, where the code exists in the conversation and never touched your disk. The agent writes it into the project directly, so there is nothing to download first.
If the reverse happens and what you have is a whole archive rather than a snippet, the structure check is different and lives in its own walkthrough: publishing a site from a ZIP.
What the tools in the search results do not give you
Paste-and-preview tools are useful, and for a throwaway look at markup they are faster than anything here. Two things they generally do not do: keep the same address when you change the code, and keep the previous version. Publishing again through the agent replaces the page at an address you already sent out, and every publish stays addressable, so undoing a bad edit is choosing the last one.
And one that matters if anyone else will look at the page: send them the version's address and they mark remarks directly on it without an account. Your agent reads those numbered threads and fixes them in the same draft; it cannot reply or resolve, so closing them stays with you.
Limits worth knowing
Static output only: HTML, CSS, browser JavaScript. No server, no database, no authentication. Taking payments on your own site is not something Birta does yet, and lead form submissions arrive in the project panel plus an email digest rather than in Slack or a webhook.
On the free tier a page gets a Birta subdomain and keeps the "made in Birta" badge; a custom domain needs Solo or Pro.
Turn the code you have into an address
One prompt publishes it, and the next version replaces it at the same address.
