Claude Live Artifacts, and the question of sharing them.

What are Claude Live Artifacts?

A normal Claude artifact renders once. You ask for a dashboard, Claude generates the HTML, and that markup is a snapshot of the data at the moment it was built. A Live Artifact is the version that does not go stale. Anthropic describes it as a persistent, interactive HTML page Claude creates for you in Cowork, connected to your apps and files, that pulls fresh data when you open it.

The connection runs over MCP. A Live Artifact is wired to the connectors you have approved, Gmail, Notion, and the rest, and reads from them directly rather than from a frozen copy baked into the page. The typical targets are dashboards, trackers, and reference tools: the kind of thing whose whole point is to reflect the current state of something, where a one-time render would be worthless an hour later.

One detail to state plainly, because it shapes everything below: a Live Artifact runs your approved connectors without asking each time. Anthropic notes that, unlike a normal session, an artifact does not prompt for permission before using a connector once it has been approved during creation. That convenience is exactly why the feature stays personal. A page that silently reads your Gmail is not a page you would hand to someone else.

How they refresh

The refresh model is the core mechanic. When you open a Live Artifact, it does not just re-render the last data it had; it re-queries the connected apps on its own. Anthropic documents a short cache that holds recent data so the artifact loads quickly, after which it goes back to the source. There is also a manual refresh button in the artifact's header for when you want to force the latest read.

That is the line between a Live Artifact and a static one. A static artifact is HTML with the numbers welded in. A Live Artifact is HTML plus a standing connection to where the numbers actually live, so the same page shows you today's figures today and tomorrow's tomorrow, without Claude regenerating it.

Who can see Claude Live Artifacts?

This is where most of the questions land, so here it is directly: at launch, the answer to who can see a Live Artifact is you, on the device where it was made.

Anthropic documents them as personal and local. There are no access controls to configure, because there is nobody else in the picture. And they live on your computer: the docs note that if you switch devices, your Live Artifacts do not come with you. The page is tied to the machine and the account that created it.

On availability, Live Artifacts are a Cowork feature, and the requirements follow from that:

• Paid plans only. Per the docs, they are available on Pro, Max, Team, and Enterprise. They are not on the free plan.
• Desktop only. They run on Claude Desktop for macOS and Windows. They are not a browser feature at launch.
• Inside Cowork. The whole feature lives in Cowork, so that is the surface where you create and open them.

Can you share Live Artifacts privately?

Not at launch. Anthropic's docs say it as clearly as it can be said: Live Artifacts are for your own use at launch, and sharing is on the roadmap. There is no private link, no password share, no email invite, and no read-only viewer for a Live Artifact today.

This is a sensible default rather than an oversight. A Live Artifact reads your connectors without prompting, so a share would mean either handing someone a page that uses your access, or rebuilding the data layer to run against theirs. Both are real design problems, and the safe launch position is to keep the artifact personal until they are solved. Sharing being on the roadmap signals Anthropic intends to address it; it just is not available now.

So if your need is a live, multi-viewer dashboard inside Claude, the honest answer is to wait for the roadmap. But many requests that get phrased as "share my Live Artifact" are actually a different need: send a fixed snapshot of an agent-made page to someone on a private link. That one has an answer today.

The gap: a private link for agent-made HTML

Full disclosure: we build Stacktree, so read this section as us describing where we fit, not as neutral advice. The two are not substitutes, they sit on different sides of one line.

A Live Artifact is a standing connection to live data, kept personal and local on purpose. What it does not do at launch is give a page a URL someone else can open. Stacktree does exactly that, and only that: it takes agent-made HTML and publishes it to a private, unguessable URL, where the URL is the credential, so the viewer needs no account and does not have to belong to any workspace. You can layer on a password, an email-domain gate with magic-link verification, or an expiry. An agent can publish the page directly over MCP from Claude, and update_site replaces it in place at one stable URL as you iterate.

The trade is the data model. Stacktree hosts static HTML; it does not hold a live connection to your apps, so a page you publish is a snapshot, not a self-refreshing dashboard. If what you want is a tracker that re-queries Gmail every time you open it, that is the Live Artifact's job and Stacktree does not replace it. If what you want is to take the dashboard Claude just built and send it to a colleague, a client, or your own phone, on a link that is private by default and not indexed, that is the gap Stacktree fills while sharing is still on the roadmap. One is live and personal; the other is a snapshot you can share. They line up rather than compete.

There is a second reason this matters: indexing. Claude's separate native Publish feature creates a public link, with no expiry, no password, and pages that may be indexed by search engines. That is the right tool for something you want the world to see, and the wrong one for an internal dashboard. A private, unguessable URL is the option in between a personal-only Live Artifact and a fully public Publish link.

What to watch

This reflects Anthropic's documentation as of June 2026, and a launch-stage feature is exactly the kind that moves. A few things would change the answers above:

1. Sharing landing. Anthropic says sharing is on the roadmap. When it ships, the most-asked question here changes, and the shape of any private-share, what gating it offers and whose connectors it uses, will be the thing to read closely.
2. Cross-device sync. Live Artifacts are local today. If they start syncing across devices, the "stays on one machine" limitation eases even before sharing arrives.
3. Plan or surface availability. Today it is paid plans, desktop, inside Cowork. Any move to the browser, or to lower tiers, widens who can build one.
4. Connector permission model. The run-without-asking behavior is what keeps artifacts personal. A change there would likely come paired with whatever sharing model Anthropic chooses.

If any of these land, the answers above will need updating. For now, this is the documented state of the feature.