Claude Code Artifacts can now pull live data through MCP connectors every time someone opens them

Claude Code Artifacts launched in beta in June 2026 with a clear premise: when you finish a coding session, you can hand someone the result as a browser-ready HTML page on a claude.ai URL, not a wall of code or a status update. That worked well for static outputs. The obvious limitation was that any data Claude pulled while building the page was baked in at creation time. A dashboard was really a screenshot dressed up as a webpage.

That changes with the update shipping in the week of July 13, 2026. Artifacts can now call MCP connectors live, every time someone opens the page. A few other shipping features round out the update: public sharing links, editor roles for Team and Enterprise plans, and the ability to create artifacts from Claude Tag sessions in Slack.

Here is what each of these means in practice.

Live data through MCP connectors

The core shift is that a published artifact is no longer a static snapshot. It is a durable shell that fetches fresh data on load, on a refresh interval, or when a viewer hits a refresh control on the page.

The creator declares which connectors the page may call as part of the publishing step. The page can only call connectors named in that declaration, so there is no open-ended data access. When someone views the page, each connector call runs through that viewer’s own claude.ai account connections, not the creator’s. Two people opening the same dashboard can see different data based on what their accounts can access.

Viewers have to approve connector access before the page makes its first call, which keeps things transparent. Responses are cached in the browser, so a page that is closed and reopened renders from the cached data immediately and then updates with a fresh call in the background.

One important restriction: artifacts that call connectors cannot be shared publicly on any plan. Public links work with no sign-in required, which means there is no authenticated viewer to broker the OAuth flow through. Connector-backed artifacts are limited to org sharing on Team and Enterprise plans, where every viewer signs in as an org member.

What this means for you: If you have been building dashboards or status pages with Claude Code and refreshing them manually whenever the underlying data changes, you can now build the page once and let it stay current. A prompt like “Build a dashboard artifact of open pull requests that pulls the live list through my GitHub connector when the page loads” is enough to get started, provided you are on v2.1.209 or later.

Artifacts now support public sharing links. Anyone with the link can open the page with no claude.ai sign-in required.

On Pro and Max plans, a public link is the only sharing method available. On Team and Enterprise plans, public sharing is off by default. An Owner has to enable it explicitly under Settings > Claude Code > Capabilities > External sharing before members can publish artifacts to public links.

This is a deliberate default. Team and Enterprise plans are scoped to keep work inside the organisation unless an admin actively opens it up.

Editor roles for Team and Enterprise

Until now, an artifact was something one person published and others could only view. Team and Enterprise plans now support editor roles, which adds a lightweight collaborative layer.

In the share dialog, you add a person and switch their role from viewer to editor. An editor can publish new versions of the artifact by giving Claude the artifact’s URL in their own session. Claude pulls the current content and republishes with their changes. It is not real-time co-editing, but it does mean the artifact can evolve without the original creator being the only person who can update it.

Audit logs capture claude_artifact_* publish and share events, so there is a record of who changed what on Enterprise plans.

Artifacts from Claude Tag sessions

Team and Enterprise users who have Claude available in Slack via Claude Tag can now generate artifacts directly from those sessions. If a conversation in Slack produces something worth sharing as a standalone page, you can now turn it into an artifact without switching context to a full Claude Code session.

Admin controls worth knowing

Connector calls from artifacts have their own toggle in Settings, separate from the general Artifacts toggle. On Enterprise plans with role-based access control, you can additionally scope artifacts to specific roles. The path is Settings > Capabilities > Enable artifact connectors.

If your organisation has not upgraded to Claude Code v2.1.209 yet, Claude will still publish artifacts, but they will use the data from the building session with no live refresh.

The bigger picture

Artifacts were always positioned as a handoff format: the output of agentic work as something you open in a browser. Live connector data extends that significantly. The page is no longer a record of what Claude found; it is a view into a system that updates on its own schedule.

The constraint that keeps this practical is the architecture. An artifact is one self-contained page with no backend. It cannot store form input, serve multiple routes, or call connectors that are not explicitly declared at publish time. It is a shareable window into your data, not a full application. That boundary is what makes it safe enough to share across an organisation without creating a sprawling set of permissions to manage.

For teams that already use Claude Code as part of their workflow, the combination of live data, editor roles, and org-level sharing makes artifacts meaningfully more useful as an internal tool. The public sharing option makes them useful for anything you want to hand to someone outside the organisation without asking them to create an account.

Updates to this story

28 July 2026: Claude’s MCP connector directory now exceeds 950 servers as new 2026-07-28 spec features roll out

Claude’s connector directory has crossed 950 listed MCP servers, used by millions of people daily, as Anthropic rolls out support for the new MCP 2026-07-28 specification across Claude products.

The new spec extends what was covered in the original article. The live-data capability introduced with Code Artifacts’ MCP connector support now has a more capable foundation beneath it: a stateless protocol core, stricter OAuth and OIDC authorization, and versioned extensions for Apps and Tasks. MCP Apps, one of those extensions, lets connectors render interactive UI directly inside the conversation, so users can work with live data inline without leaving Claude.

Two additions are particularly relevant for teams. Enterprise-managed auth lets admins provision connectors organization-wide through their identity provider, so users inherit access on first login with no per-user OAuth prompts. MCP Tunnels (research preview) connect Claude to MCP servers inside private networks without public endpoints or firewall changes.

Connector developers should note the stateless core is a breaking change: session IDs are removed from the wire and OAuth requirements are materially stricter. Servers that do not migrate will degrade and stop working. The MCP migration guide at mcpmigrate.dev covers what needs to change. Rollout across Claude product surfaces is ongoing.