Table of Contents
On June 18, 2026, Anthropic shipped Claude Code Artifacts (beta) — a feature that turns a terminal coding session into a "live" web page your team can share. Instead of Claude Code streaming endless git diff and logs as text, it can publish an annotated PR walkthrough, a self-updating dashboard, or an incident timeline as a single page.
This article covers what Artifacts is, how it differs from 2024's claude.ai Artifacts, how to use it, its limits, and availability — based on the official docs. Three takeaways up front. (1) It publishes a session's output as one page at a private claude.ai URL, and the content updates as the work progresses. (2) There is no dedicated /artifact command — you ask in plain language ("make a page that walks through this PR with annotations"). (3) It's available on Pro, Max, Team, and Enterprise plans, usable only in sessions signed in via /login (not with an API key).
Turn a session into a "shared page"
— from terminal output to a live page you can look at and interact with
It publishes to a private claude.ai URL (visible only to you until you share it).
Each update becomes a new version at the same URL, and the page refreshes in place for whoever has it open.
1. What Claude Code Artifacts is
The official definition: "an artifact is a live, interactive web page that Claude Code publishes from your session to a private URL on claude.ai." Open it in a browser and it updates in place as the session continues. The page is "built from anything your session can reach, including your codebase and data it pulls through your connected tools" (MCP connectors).
The point is to "turn output that's hard to convey as terminal text into one page you can look at and interact with." For example, during a long investigation you can share an incident timeline that fills in as you work, or a release checklist that checks itself off as tasks complete. Anthropic positions it as collaboration tooling that "lets teams spend less time on status updates and more time building." Think of it as a new output surface for Claude Code, alongside Skills and MCP.
2. How it differs from 2024's claude.ai Artifacts
Many people will think of the side-panel (canvas) Artifacts that appeared in the claude.ai chat in 2024. They share a name and lineage, but Claude Code Artifacts is a different surface. Since the two are easy to confuse, here's the breakdown (note: this contrast is our own synthesis — Anthropic does not present them side by side).
| claude.ai Artifacts (2024) | Claude Code Artifacts (2026) | |
|---|---|---|
| Where it shows | A canvas beside the chat in claude.ai | A standalone page from a terminal coding session |
| Source of content | The chat conversation | The whole session (codebase, connectors) |
| Live-updating for viewers | No (you regenerate in chat) | Yes (published page refreshes in place) |
| Sharing scope | Can be made public | Private by default; share within your org or via a public link (on Team / Enterprise, an Owner must enable public sharing) |
In one line: the 2024 version "renders an artifact inside the chat," while the 2026 Claude Code version "turns a terminal coding session into a self-updating, shareable web page." Same name, but different use case and a different way of sharing.
3. What it's good for
The use cases Anthropic names are exactly the kind where "seeing a page beats reading text."
When a "page" beats text
The common test: "is this output easier to look at and interact with than to read line by line?"
4. How to use it
The biggest quirk: there is no dedicated /artifact command. You ask in plain language, or Claude offers one when the output suits a page. The flow:
# 1) Ask in plain language during a session (examples)
Make an artifact that walks through this PR with the diff annotated inline.
Build a dashboard artifact of last week's deploy failures by service and keep it updated as you investigate.
# 2) Claude writes a .html (or .htm / .md) — to a temp folder outside your project unless you name a location — and publishes it
# In plan mode, or when the page declares capabilities such as connector calls, Claude asks first
# 3) The URL is printed and opens in your browser. Reopen the latest with Ctrl+]
# Disable auto-open with CLAUDE_CODE_ARTIFACT_AUTO_OPEN=0To update, just ask Claude to revise — it edits the file and publishes to the same URL, and each publish becomes a new version (the Share control lets you choose which version viewers see). To update from a different session, paste the artifact's URL (otherwise a new artifact is created). A new artifact is visible only to you at first. Share it from the Share control in the page header: on Team / Enterprise you can grant specific people or your whole org, and a public link lets anyone open it without signing in to claude.ai (on Team / Enterprise, only if an Owner has enabled external sharing). People you share with are viewers by default; on Team / Enterprise you can also make someone an editor. All your artifacts are listed in the gallery at claude.ai/code/artifacts.
5. Limits — it's not an "app"
The official docs state plainly: "an artifact is a capture of work, not an application." Misread that and you'll be disappointed, so note the constraints.
Main constraints (official)
· No backend: a single static page. It can't authenticate viewers or serve multiple routes (in-page anchors only). To pull fresh data at view time, it calls MCP connectors through claude.ai.
· Limited external requests: under a strict CSP, the page can load only typefaces from Google Fonts and scripts from five public CDN hosts (cdnjs, unpkg, jsDelivr, etc.). Other external scripts/CSS/fonts/images are blocked, and fetch, XHR, and WebSocket reach only the page's own origin and Google Fonts. Other CSS/JS is inlined; images are embedded as data URIs.
· File types and size: only .html / .htm / .md. Rendered page must be ≤ 16 MiB (large images are the usual cause of failed publishes).
· Token cost: inlined CSS/JS and data-URI images mean it consumes more tokens than terminal text.
So Artifacts is not "a way to build a running app" — it's "a way to solidify a session's output into one shareable, readable page." If you need a real tool that talks to external services, go to normal development and deploy. To show it outside your org, share a public link (on Team / Enterprise, if an Owner has enabled external sharing), or have Claude output the raw HTML file.
6. Availability and admin controls
Requirements as of October 4, 2026 (confirm the latest in the docs).
| Item | Details |
|---|---|
| Plans | Pro / Max / Team / Enterprise. On by default on Team / Enterprise; an Owner can turn it off for the org |
| Status | Announced as a beta on June 18, 2026 (initially Team / Enterprise only) |
| Where it works | Claude Code CLI, or the Claude desktop app (v1.13576.0+). Off by default in Agent SDK / GitHub Action / MCP-server contexts |
| Auth | Must be signed in to claude.ai via /login. API keys, gateway tokens, and cloud credentials cannot publish |
| Model provider | Anthropic API only (not Amazon Bedrock / Google Vertex / Microsoft Foundry) |
| Org policy | Disabled for orgs with CMEK, HIPAA, or Zero Data Retention (ZDR) |
For admins there are org-level on/off, role-based scoping (Enterprise RBAC), retention policies, an audit log (claude_artifact_*), and a Compliance API. To disable it yourself, use "enableArtifact": false in settings, the env var CLAUDE_CODE_DISABLE_ARTIFACT=1, or deny the Artifact permission. If you get stuck on sign-in, see auth / login error fixes.
Summary
Claude Code Artifacts (announced June 18, 2026) turns a terminal coding session into a "live" shared page at a private claude.ai URL. PR walkthroughs, dashboards, incident timelines, checklists — output that's easier to see than to read — shared with your team. It's distinct from 2024's canvas-style claude.ai Artifacts: session-sourced, self-updating, private by default. There's no /artifact command — you ask in plain language, Claude writes a .html and publishes, and each update is a new version at the same URL.
But it's "a capture of work, not an app" — no backend, limited external requests (CSP), a single page, ≤ 16 MiB. It works on Pro / Max / Team / Enterprise plans and requires /login sign-in (not API keys) on the Anthropic API only (not Bedrock, etc.). For teams that want to cut the cost of status updates, it's a strong option. Related: Claude Agent Skills, MCP, Claude Cowork.
FAQ
Q. Is this the same as the 2024 claude.ai Artifacts?
A. No, it's different. The 2024 version renders an artifact in a canvas beside the chat in claude.ai. Claude Code Artifacts turns a terminal coding session into a standalone page at a private claude.ai URL, built from the whole session (codebase, connectors), self-updating, and private by default. They share a name and lineage, but the surface and the way of sharing differ.
Q. How do I create one? Is there a command?
A. There is no dedicated /artifact command. Ask in plain language during a session ("make an artifact that…"), or Claude offers one when the output suits a page. Claude writes a .html (or .htm / .md), publishes it, and prints a URL (in plan mode, or when the page declares capabilities such as connector calls, it asks first). Reopen the latest with Ctrl+].
Q. Can anyone view the page? Can I make it public?
A. A new artifact is visible only to you at first. Use Share in the page header: on Team / Enterprise you can share with specific people or the whole org. A public link lets people open it without signing in to claude.ai (on Team / Enterprise, only if an Owner has enabled external sharing). People you share with are viewers by default; on Team / Enterprise you can also make someone an editor.
Q. Can I build a working web app?
A. No. The docs call an artifact "a capture of work." It has no backend, limited external requests (the CSP lets fetch/XHR/WebSocket reach only the page's own origin and a few hosts), a single page, and a ≤ 16 MiB limit. If you need arbitrary API calls or data storage, build and deploy normally. Artifacts is specialized for "sharing output in a readable form."
Q. Can I use it on my plan?
A. As of October 4, 2026, it's available on Pro / Max / Team / Enterprise plans. It's on by default on Team / Enterprise, and an Owner can turn it off for the org. You need a session signed in to claude.ai via /login, and API keys / cloud credentials cannot publish. The provider is the Anthropic API only (not Bedrock/Vertex/Foundry). Scope can change, so confirm the latest in the official docs.
