Anthropic's Claude Code Turns AI Sessions Into Live Shareable Dashboards

Claude Code just got a meaningful upgrade to its artifacts feature. Anthropic has announced that Claude Code artifacts can now turn a session's work into a live, interactive, shareable HTML page that connects to multiple data sources and surfaces on a URL teammates can open instantly. The new twist: those artifacts can now call MCP connectors at view time, meaning the data refreshes for each viewer on demand using their own credentials.

The problem with static snapshots

Before this update, Claude Code artifacts were already useful for sharing session output. Claude would build an HTML artifact, connect it to MCP servers, and save it to the Live Artifacts tab in the sidebar, where a toggle enabled real-time data refresh. But the model was creator-centric: the person who built the artifact had to re-run their session to get fresh numbers, and sharing meant sharing a frozen snapshot.

The new capability flips that model entirely. Users authenticate with their Claude account and interact with their own instance of the artifact, with no API keys required and no costs to the creator. Each viewer's MCP connections supply the data, so a dashboard showing Jira tickets or Slack messages shows their Jira tickets and Slack messages, not the creator's.

How it actually works

MCP, or Model Context Protocol, is an open standard that lets Claude connect to external tools and data sources through a standardized interface. Think of it as a universal adapter layer. Claude Code can connect to hundreds of external tools and data sources through MCP. There are now 554 tracked MCP integrations across Anthropic's connector directory, organized across 30 categories.

MCP integration for artifacts is available on Pro, Max, Team, and Enterprise plans on Claude web and desktop. Artifacts can connect to external services through MCP, enabling interactive applications that read from and write to tools like Asana, Google Calendar, and Slack.

Each user must authenticate MCP servers independently, even when using shared or published artifacts. Organization admins can enable or disable artifact MCP access at the organization level but cannot manage which specific MCP servers artifacts can use. This is the key security property: the artifact itself never holds credentials, and viewers can only see data they already have access to.

What you can build

The practical surface area here is large. Common use cases include real-time dashboards pulling metrics from a database or analytics API on each load, status pages showing current system health, analytics views displaying live campaign or sales data without manual refresh, and live data tables querying a database and rendering results as a formatted table.

The Claude Code artifact workflow looks roughly like this:

1. Prompt Claude Code to build a dashboard or interactive page, naming the MCP-connected data sources to use.
2. Claude writes the HTML page, wires in the MCP data bindings, and publishes it to a private URL on claude.ai.
3. Share the link within your organization. Each viewer authenticates their own MCP connectors on first interaction.
4. The page updates in place as the session continues, with each publish creating a new version in the history.

Here is a minimal prompt pattern that works well:

Build a dashboard artifact showing open GitHub issues for this repo,
recent Sentry errors from the last 24h, and today's deploy status.
Keep it updated as you investigate. Use the GitHub and Sentry MCP connectors.

The limits are real

This is not a replacement for a proper BI tool or internal application. Anthropic's documentation is explicit: "An artifact is a capture of work, not an application." Each artifact is a single, self-contained HTML page capped at a rendered size of 16 MiB.

The page constraints matter for anyone planning to build on this:

• No external requests at render time -- the Content Security Policy blocks fetch, XHR, and WebSocket calls. MCP data is pulled through Claude's infrastructure, not directly from the page.
• No backend -- the artifact cannot store form submissions, authenticate viewers itself, or serve multiple routes.
• Not available on publicly shared artifacts -- MCP-connected artifacts require authenticated viewers, so you cannot publish one to a public link and have it pull live data for anonymous visitors.
• No CMEK, HIPAA, or Zero Data Retention orgs -- these compliance configurations are incompatible with the artifacts feature entirely.

Claude artifacts are not a replacement for Tableau, Looker, or Power BI. Those platforms offer role-based access controls, embedded analytics, audit logs, and enterprise governance that Claude does not have yet. If your organization needs org-wide monitoring or has strict data compliance requirements, a dedicated BI tool is still the right call.

Where this fits in the bigger picture

Anthropic's update comes more than two weeks after OpenAI released a major update to its own Codex platform, introducing a similar enterprise hosting feature called "Sites." OpenAI Sites supports persistent backend infrastructure, including relational databases and object storage, and allows public sign-ins with external identity providers. Claude's approach is deliberately lighter: no backend, no persistent state, but faster to build and zero infrastructure to manage.

The per-viewer MCP model is the more interesting architectural bet. Rather than the creator pre-fetching data and baking it into the page, the artifact becomes a template that each viewer hydrates with their own connected data. This is the architectural moment when Claude becomes a hub -- not a tool running alongside your stack, but a layer that reaches into it and acts on your behalf. A single artifact can serve an entire team, with each person seeing only what they have access to, without the creator managing any of that access control.

Artifacts in Claude Code are available on Team and Enterprise plans. Usage counts against each viewer's own Claude subscription limits, not the creator's. For Team and Enterprise plans, when you share AI-powered artifacts within your organization, team members can use them without incurring additional costs to the creator.