Sharing a private HTML report is a solved problem. Choosing the right tool isn't.
A dozen tools turn an AI-made HTML file into a link only the right people can open. They are not interchangeable — each is optimizing for something different. This guide sorts them by what you actually care about, and states plainly where each one wins.
The field, at a glance
Start here
Find the situation that sounds like yours.
Just needs to be seen
It isn't sensitive — a demo, a one-pager, a portfolio piece. You want the shortest path to a link that renders.
Anthropic Publish · ShareDuo
It has to stay private
A leak would cost you. Either prove exactly who opened it, or keep the host from being able to read it at all.
Peony · Stacktree
One client, over months
You'll hand the same person work again and again, and want it collected in one durable place — not scattered across links and inboxes.
PageVault · SuiteDash
The doc is a conversation
You want comments, reactions and iteration on the artifact itself, not a one-way hand-off.
jonesphillip/sharehtml · HTML Share
You own the whole stack
Free, self-hosted, on your own domain, open enough to read the source. No vendor sitting in the middle.
PageVault · jonesphillip/sharehtml
Most of these overlap on the basics — a link, a password, HTML that renders. The differences that decide it are below.
Compare
Put any two side by side.
Tool A
Tool B
✓ yes~ partial✕ nohighlight = leads that row★ effectively unique to PageVault
Capability
A
B
The map
Change what matters. Watch the leader change.
What are you optimizing for?
AI-nativemanual
the open lane
——
—
—
Tap any point to read what it is.
What's changing
The place you drop the file has started rendering it.
Every tool above exists because a recipient opened an HTML file and saw markup. That premise is eroding — unevenly, and only inside the walls of whatever app you dropped it in.
Slack
Both, inline. Attached HTML started rendering as a page in mid-June 2026, and Markdown files now render too. Neither is documented — it is a gradual rollout, so the same file can render in one workspace and download in another. Canvas has handled Markdown all along, and Slackbot became an MCP client in June, so a publish tool can be called from the channel itself.
OneDrive · SharePoint
Markdown, natively — view, edit and split-pane preview, generally available April 2026. HTML, no.
Box
Markdown, natively — a built-in editor with preview and split modes since May 2026. HTML previews as code.
Google Drive
Neither. A .md file is raw text and an .html file is markup or a download — which is why there is now a cottage industry of extensions selling the missing preview.
Dropbox
Neither. Two hundred–odd previewable formats, and your Markdown is still hashes and asterisks.
Two things follow. Markdown is quietly becoming a first-class citizen nearly everywhere; interactive HTML still is not, except in Slack, which now renders both. And none of it produces an address — a rendered file in a channel or a drive has no URL a client outside it can open, no gate you set, and no place it accumulates. Rendering is a viewer. That is a different job from sharing, and a different job again from keeping.
Listed, not assessed. This section will be the first part of this page to go stale.
Where we fit
This guide was made by PageVault. Here's our hand, face up.
On the map above, PageVault leads two frames — the collection and artifact handling — and sits mid-pack or lower on every other one. We're not the most private (Stacktree is), the most accountable (Peony is), or the most collaborative (sharehtml is). That's fine. We're aiming at a different square.
Almost every other tool treats the link as the unit: publish a file, get a URL, done. PageVault treats the client as the unit — every artifact lands in one durable portal, gated to their email, that they bookmark once. No single capability below is ours alone anymore; data rooms have collections, five tools have an MCP server. What's rare is the shape of the whole thing:
What's genuinely rare here
The whole combination. Free, self-hosted on your own domain, renders the artifact and Markdown, keeps it as a per-client collection, open-source enough to read in a sitting. Any one piece you can find elsewhere; the set is the gap.
Your AI reads the collection back. An MCP server lets Claude search and read a client's whole corpus later — "what did we decide about the CDC migration in March?" The nearest rival, Peony, has a bidirectional MCP and a free tier we were wrong to gloss over — but it reads a deal room, which is a different object from a running record of what you handed a client.
When to skip PageVault
The artifact is confidential. Watermarks, NDA gates, a verified page-by-page audit trail → Peony.
The host must never see plaintext → Stacktree's client-side encryption.
You want live feedback on the doc → jonesphillip/sharehtml.
You need a CRM, invoicing, e-sign → SuiteDash or Copilot.
You have under ~5 clients. A shared folder per client is genuinely faster until the artifacts pile up.
Free, MIT-licensed, self-hosted on Cloudflare's free tier. Publishing over MCP works the same from Claude Code and the hosted chat surfaces — claude.ai, Claude Desktop and mobile all connect over OAuth. danjamk.github.io/pagevault → · source on GitHub