Why Claude's "Publish" Button Doesn't Work for Internal Sharing

What Claude's Publish button does

When you create an artifact in Claude's web UI, there's a Publish button. Clicking it does one thing: creates a publicly accessible URL on claude.ai/public/artifacts/{uuid}.

On Free, Pro, and Max plans, this URL is fully public. Non-Claude users can view the artifact without signing in. The URL is tied to the artifact, and unpublishing is permanent and irreversible, destroying all associated storage.

There's no company restriction, no domain allow-list, and no authentication. Anyone with the link can view it.

The gap: no middle ground between public and org-only

Claude's sharing model has two modes:

Team/Enterprise plans: Artifacts can be shared within the Claude organization. Viewers must have Claude seats on your plan. If your PM, VP, or sales team don't have Claude seats, they can't view it. At $25–$30/seat/month per viewer, sharing an artifact with 20 non-Claude teammates costs $500–$600/month in new seats.

Free/Pro/Max plans: The URL is fully public, with no restriction of any kind.

There is no middle option. There's no "only people with @yourcompany.com emails" setting, no domain restriction, and no one-time password for specific external viewers. The Publish button is binary: public or org-internal (with seats required).

Why this matters

Internal content is internal for a reason.

Architecture proposals contain system design decisions competitors would value. Competitive analyses contain research that shouldn't be public. Financial models contain data that's material and regulated. Client presentations reference relationships that are confidential.

Putting any of this at a publicly accessible URL (even an obscure one) doesn't satisfy security policy, compliance requirements, or basic information hygiene. "Nobody will guess the URL" is not access control; it's a 128-character obfuscation that disappears the moment someone forwards the link.

The Publish button is designed for demos and public sharing, not for internal distribution of sensitive content.

The second problem: Claude Code's publishing is Claude-org-only

Claude Code Artifacts turns session work into interactive pages (PR walkthroughs, reports, dashboards) with live updates and versions. It exists, it works, and it is org-private by default. But it solves only the native Claude org workflow:

• Sharing is Team/Enterprise-only, and viewers need a Claude seat. Publishing an artifact works on Pro, Max, Team, and Enterprise, but on Pro and Max the page is private to you alone — you can't share it with anyone. Only on Team and Enterprise can you share it, and then only with authenticated seats inside your Claude org. A PM, VP, or sales lead without a seat on your plan cannot open the page.
• Claude-only as a source. The page has to come from Claude. There is no path to publish the same artifact from Codex, Cursor, CI, or a local model.
• No external or public sharing. No guest links, no @yourcompany.com middle ground for people outside your Claude org.
• No comments. Feedback lives somewhere else (Slack, email) and never lands back on the artifact.

So the gap is narrower than "there's no Publish button" but it is the same shape as the web UI gap above: viewing is bound to Claude seats, the source is locked to Claude, and there is no neutral layer where any agent's HTML lives behind your company login.

This matters because Claude Code is one of several places AI-generated HTML comes from. Engineers also generate architecture diagrams, competitive analyses, and visual reports from Cursor, Codex, CI jobs, and local models. None of those can publish into Claude Code Artifacts, and every coding agent that writes HTML to disk still has no native path to a company-authenticated URL.

The workarounds people use (and why they fail)

Screenshot + Slack: Loses all interactivity. Charts freeze, tabs disappear, and the artifact is degraded permanently.

Download + email: HTML attachments are blocked by most corporate email security policies. Even when they arrive, opening locally breaks relative paths and network-dependent JS.

Share via Claude org: Works only if every viewer has a Claude seat. At $25–$30/seat/month per viewer, this is prohibitively expensive for wide distribution.

Tell the viewer to sign up for Claude: Adds friction and cost. Most stakeholders who need to see an artifact (PMs, VPs, legal, sales) shouldn't need an AI coding tool subscription to view a report.

The actual solution: publish outside Claude to a gated URL

For Claude Code users (CLI workflow):

claude code "Generate an interactive architecture diagram as HTML"
 
dsp publish ./architecture.html
 

Viewers click the link, authenticate with their company Google or Microsoft email, and see the full interactive page in five seconds, once.

For Claude Desktop users (no terminal):

Configure the display.dev MCP server once. Then, in any Claude conversation, say: "Publish this to my team." Claude calls the MCP publish tool and returns a company-gated URL in the conversation.

No terminal, no file download, and no configuration after initial setup.

What display.dev adds that Claude's Publish button doesn't

Claude creates the artifact, and display.dev is where you publish it for the company. These tools are complementary, not competing: Claude is an upstream creator, and display.dev is the downstream distribution layer.

Publish your first Claude artifact behind company auth, free →