What happened

Claude Code can now turn session output into interactive pages that pull current information from connected services.

Anthropic first added public sharing and collaborative editing to Artifacts. It then let those pages call Model Context Protocol connectors, which connect Claude to external tools and data.

Claude Tag can create internal Artifacts from a Slack request. A team could ask for a dashboard in a thread and receive a working page. Claude Tag is currently a beta feature for Team and Enterprise customers.

That means an Artifact no longer has to freeze the information available when Claude created it. A dashboard can fetch open pull requests from GitHub when someone loads the page. Another could show current project data from tools such as Asana, Google Calendar, or Slack.

The feature is available on Pro, Max, Team, and Enterprise plans. Connector calls require Claude Code version 2.1.209 or later, according to Anthropic’s Artifact documentation.

The viewer supplies the access

Connector-backed Artifacts use the viewer’s account, not the creator’s. Every viewer must have the required connector and approve its first call.

Two people can therefore open the same dashboard and see different results. Each sees only the information available through their own connected account.

The Artifact never receives a person’s credentials. Claude.ai makes each connector call on the page’s behalf. Controls that change external data also run with the viewer’s permissions.

Claude limits each page to the connectors declared when it is published. Local MCP servers configured through files such as .mcp.json can help build the page. The published Artifact cannot call those local servers when someone views it.

Why it matters

This gives builders a faster way to package agent output for other people. A session can produce a page instead of another long message, exported report, or pasted screenshot.

Useful examples include:

  • a release dashboard that refreshes from GitHub
  • a pull-request walkthrough with annotated code changes
  • an incident timeline that updates during an investigation
  • a status page that shows each viewer the projects they can access

This is especially useful when the result needs explanation or interaction, but not a complete application. It can shorten the distance between asking Claude for an analysis and giving someone else a usable view of it.

Collaborative editing extends that workflow. Team and Enterprise users can let another person edit an Artifact by giving Claude its URL in a separate session. Claude republishes the changes to the same page, and viewers see the new version.

An Artifact is still not a full application

Anthropic’s documentation draws a clear boundary. A Claude Code Artifact is one self-contained page without its own backend. It cannot serve multiple routes or store form submissions.

That makes it suitable for dashboards, explainers, comparisons, and temporary internal tools. It is a poor substitute for software that needs durable business logic, independent user management, or a database you control.

Sharing also has an important limit. An Artifact that calls connectors cannot use a public link. Teams can keep it private or share it inside their organization, where viewers authenticate before accessing it.

Public Artifacts remain useful for demonstrations and static interactive pages. Connected Artifacts belong on the internal side of the line because they can reach private data and take actions.

What builders should do next

Start with a page you already recreate manually. A weekly project dashboard or release summary is safer than a workflow that changes production data.

Ask Claude to name the connector required for every live section. It should also show a useful fallback when a viewer has not connected that service.

Test the Artifact with an account that has fewer permissions than yours. Confirm that the page shows only what that user can already access.

Treat every write control as a real product action. A button that updates an issue or posts a message deserves the same confirmation and error handling as a normal internal tool.

The practical opportunity is not to replace application development. It is to turn the useful output of a coding session into a live page before building more infrastructure than the job requires.

End of article