Guides to what Claude Artifacts can and cannot do
Artifacts are very good at some things and structurally incapable of others, and the difference is not always obvious until you are halfway into building something. These pages answer the specific questions people hit, quote Anthropic's documentation rather than paraphrasing it, and say plainly when the right move is to keep using Artifacts.
Last checked against Anthropic's documentation on . Claude changes often. If something here no longer matches the help centre, the help centre is right.
Can Artifacts store data?
Yes, on paid plans, 20 MB, text only, and only once published. Why storage silently does nothing while you are testing.
Unpublishing an Artifact
It kills the link, permanently deletes all stored data, and you can never publish that artifact again. How to export first.
Sharing an Artifact
Publish and send the link. Viewers need no Claude account, but Team and Enterprise artifacts stay inside the organisation.
Can other people use it?
Anyone with the link can. What they cannot be is recognised, which is what breaks anything with names, roles or private rows.
Connecting Claude to a database
Three different questions with three answers: MCP for your own database, artifact storage, or a hosted app with real accounts.
Homespun vs Artifacts
The two compared line by line, with the numbers, and a long section on the cases where an artifact is the better product and you should stay.
Hosting an artifact
Building an app with Claude Code
Deploying a Claude Code app
Claude Code to Vercel
Claude Code and databases
AppSheet alternative
Converting a Google Sheet
Airtable alternative
Notion alternative for apps
Google Sheets as a database
The short version
If you are here for one paragraph: a published Claude Artifact is a single self-contained page that anyone can open without an account. On paid plans it can persist up to 20 MB of text, but only once published, and it has no way to tell one visitor from another. That combination makes it excellent for anything a person uses alone or a crowd uses anonymously, and unable to carry anything where it matters which person did what.
An artifact is a capture of work, not an application.
Share session output as artifacts, Claude Code documentation
That sentence is Anthropic's, from their own developer documentation, and it is the most useful single line written about artifacts. Almost every question below is a consequence of it.
One qualification travels with all of it, because the simple version of the story is now wrong: an artifact is not completely sealed off from the world. It can call MCP connectors while somebody is viewing it, through that viewer's own connections rather than yours, and an artifact that does so cannot be shared to a public link on any plan. The pages below quote Anthropic on both halves rather than only the half that suits us.
Who wrote this, and the bias you should assume
Homespun hosts apps that agents build, so we have an obvious interest in the moment somebody outgrows an artifact. We have tried to earn the read anyway: every claim about Claude here is quoted from Anthropic's own help centre and linked, every page has a section on where Artifacts genuinely beat us, and none of them ends with a demo request. If a page reads like a pitch rather than a reference, it has failed and we would rather fix it.
Want the other side of it? What Homespun does, and how to connect it to your AI.