Claude Code Artifacts Explained: Who Gets It Now, How It Works, and What It’s Actually For

Last updated: July 4, 2026

Claude Code Artifacts launched in beta on June 18, 2026 — and most coverage you’ll find is already outdated. The big change: as of early July, Artifacts is no longer Team/Enterprise-only. Pro and Max plans can publish them too (with one important catch around sharing). Here’s the current state, tested.

TL;DR: Any paid Claude Code plan can now ask Claude to publish an artifact — a live, private web page on claude.ai built from your session. Team/Enterprise users can share them; Pro/Max artifacts stay private to your account. It’s a capture of work, not an app: 16 MiB max, no backend, no external requests.

What Claude Code Artifacts actually is (and isn’t)

Per Anthropic’s docs, an artifact is “a live, interactive web page that Claude Code publishes from your session to a private URL on claude.ai.” It’s built from your full session context — the codebase, connected MCP servers, and the conversation itself — and it updates in place as the session continues.

Think: a PR walkthrough with the diff annotated inline, a dashboard summarizing deploy failures, a release checklist that updates itself as your session works through it.

What it is not: a deployment tool. Anthropic’s own framing is that an artifact is “a capture of work, not an application.” No backend, no data storage, no auth for viewers, no API calls at view time.

Who gets Artifacts right now: plan-by-plan

PlanPublish?Share?
Free
Pro / Max✅ (since ~July 2)❌ — private to your account only
Team✅ (on by default)✅ specific people or whole org
Enterprise✅ (admin must enable)✅ + role-based control

Details that matter:

  • Enterprise: an Owner has to flip the toggle at Settings → Claude Code → Capabilities → Artifacts in the claude.ai admin console. Roles can be scoped further under Settings → Roles.
  • Surfaces: the CLI, or the Claude desktop app v1.13576.0+. Pages open in any browser.
  • You must be signed in via /login to claude.ai. Sessions authenticated with an API key, gateway token, or cloud-provider credentials cannot publish.
  • Anthropic API only — not available on Bedrock, Vertex AI, or Microsoft Foundry. CMEK, HIPAA, and Zero Data Retention orgs are also excluded.

⚠️ If you read a guide written before July that says “Team and Enterprise only” — that was true at launch and isn’t anymore.

Creating your first artifact: a walkthrough

There’s no slash command — it’s prompt-driven. In a session:

Make an artifact that walks through this PR with the diff annotated inline.

What happens next:

  1. Claude writes an .html (or .md) file into your project.
  2. Before the first publish, it asks permission: Claude wants to publish “…” to a private page on claude.ai. Republishing later doesn’t re-prompt.
  3. Claude prints the URL and your browser opens it automatically (turn that off with CLAUDE_CODE_ARTIFACT_AUTO_OPEN=0). Ctrl+] reopens the latest artifact from the terminal.

To update it, just keep working and ask Claude to revise — same URL, new version. One gotcha: cross-session updates need the artifact URL pasted in, otherwise a new session creates a brand-new artifact.

Claude Code building and publishing an artifact from the terminal
Published Claude Code artifact page with navigation rail and service mapping table

Claude also honors your design system: design tokens in CLAUDE.md or theme files take precedence over its defaults (your explicit prompt beats both).

The limits you’ll hit: 16 MiB, CSP, no backend

These are hard constraints, straight from the docs:

  • 16 MiB rendered page cap. Oversized embedded images are the usual publish-failure cause — everything (CSS, JS, images as data URIs) is inlined into one page.
  • Strict CSP: no external scripts, stylesheets, fonts, or images; fetch, XHR, and WebSockets are blocked. The page cannot phone home.
  • No backend: forms don’t store input, nothing authenticates viewers, no view-time API calls.
  • Single page: relative links don’t resolve — multi-section content means in-page anchors.
  • Viewers load pages from a sandboxed *.claudeusercontent.com origin; locked-down corporate networks need that allowlisted alongside claude.ai.

If a publish silently fails, check the size first — and know that when requirements aren’t met, the feature degrades to just writing a local HTML file.

Version history, gallery, and sharing

  • Every publish creates a new version at the same URL; you can restore any prior version, and a share control decides whether viewers see a pinned version or “always latest.”
  • Your gallery lives at claude.ai/code/artifacts.
  • Sharing (Team/Enterprise only): specific people or your whole org, view-only. There are no public links, period — viewers must be signed-in members of your org. Need to send one outside? Ask Claude for the raw HTML file instead.
Claude Code artifacts gallery at claude.ai/code/artifacts

Does it cost extra?

No separate fee — but it’s not free in tokens. Official docs: “Generating an artifact uses output tokens like any other response, and a styled page is more token-intensive than the same content as terminal text.” The drivers are inline CSS/JS and data-URI images. If you’re watching your usage limits, prefer SVG over raster images and skip interactivity you don’t need.

How this differs from claude.ai chat Artifacts

Same name, different animal:

Chat artifacts (claude.ai)Claude Code artifacts
Since2024June 2026 (beta)
PlansAll, incl. FreePaid plans only
Built fromThe chatYour repo + MCP + session
SharingPublic links possibleOrg-only, never public
Dynamic featuresCan call AI, 20 MB storageFully static

Artifacts vs just pushing a preview deploy

Honest take: if you already have Vercel/Netlify previews wired up, artifacts won’t replace them — previews run your actual app. Where artifacts win is the explanation layer: things you’d otherwise write in a wall of Slack text. A PR walkthrough for a reviewer who doesn’t want to check out the branch. A “what broke in this deploy” summary pulling from your MCP-connected monitoring. A live checklist during a long refactor session.

The tell: if the page needs to do something, deploy it. If it needs to show something from a working session, artifact it.

FAQ

Is there a slash command? No — describe what you want in plain language, or Claude may offer one proactively.

Can I turn it off? Yes: "disableArtifact": true in settings, CLAUDE_CODE_DISABLE_ARTIFACT=1, or add Artifact to permissions.deny.

Can free users view a shared artifact? Sharing is org-membership-based (Team/Enterprise). Pro/Max artifacts can’t be shared at all yet.

Does it work in the VS Code extension or Claude Code on the web? Publishing is documented for the CLI and desktop app (v1.13576.0+). Pages themselves open in any browser.


WorkflowDen is an independent site, not affiliated with or endorsed by Anthropic, PBC. Tested on my own machine — version numbers visible in screenshots.