claude code week 29: artifacts go live with mcp — dashboards that pull real data on view
Claude Code v2.1.213–219 (week 29, July 13–17) shipped a capability that changes how you share work: published artifacts can now call MCP connectors every time someone opens them, so a dashboard you built in a session stays current rather than freezing at the moment you built it. Week 29 also added public sharing links, collaborative editor roles, the /fork command, and screen reader support.
The most interesting thing in Claude Code's week 29 release (July 13–17, 2026, versions v2.1.213–219) isn't a new command — it's a change to what a published artifact actually is.
Previously, when you built a dashboard or report inside a Claude Code session and shared it, the recipient got a snapshot: whatever data the artifact contained at the moment you published it. If the underlying data changed, the published version didn't know.
Week 29 changes that. Published artifacts can now call MCP connectors each time someone views them. The dashboard you built in your session can pull live data when a teammate opens it, take actions on demand, and stay current without you rebuilding and republishing.
how the permission model works
The permission design is worth understanding because it's not obvious. Each MCP call runs through the viewing account's own connections — not yours. The viewer approves access before the page's first connector call fires. That means the artifact author doesn't get ambient access to the viewer's data sources, and the viewer has explicit control over what the artifact is allowed to reach.
This is a meaningful architectural decision. The alternative — baking the author's credentials into the artifact — would be convenient but would create a significant blast radius if a shared artifact landed in the wrong hands. The viewer-approval model is the right tradeoff for a feature designed for team use.
collaboration features: sharing links and editor roles
Week 29 also shipped two collaboration primitives that were missing before:
Public sharing linkslet you send a Claude Code artifact to anyone — not just teammates on the same plan. This is the low-friction path for sharing a report with a stakeholder who doesn't have a seat.
Editor roleson Team and Enterprise plans let you give a collaborator write access to a shared artifact rather than read-only access. Previously all sharing was effectively read-only from the collaborator's side.
A third addition: artifacts created from Claude Tag sessions (Claude's web-accessible interface) can now be saved and shared through the same artifact infrastructure as desktop Claude Code sessions.
/fork: branch a conversation without interrupting it
The /forkcommand ships in week 29 and solves a specific problem: you're mid-session, you want to explore a different direction, but you don't want to diverge from the current conversation or lose the context you've built up.
/fork copies your current conversation into a new background session while you keep working in the original. The forked session has the full context of everything up to that point, and the two sessions run independently from there. You can use /resume to switch back and forth.
The obvious use case is exploring an alternative implementation approach when you're not sure which direction is right. Fork, try it in the background, come back and compare.
auto mode now works on Bedrock, GCP, and Microsoft Foundry without an opt-in variable
Historically, using Claude Code's auto mode on Amazon Bedrock, Google Cloud's Agent Platform, and Microsoft Foundry required setting an environment variable to opt in. Week 29 removes that requirement — auto mode activates on all three platforms without any additional configuration.
This matters for teams deploying Claude Code in enterprise cloud environments where setting custom environment variables through managed deployment pipelines adds operational overhead.
screen reader support
Week 29 adds a screen reader mode that replaces the visual terminal interface with plain, linear text compatible with VoiceOver (macOS) and NVDA (Windows). This is a gap that's existed since Claude Code launched — the rich terminal UI that makes the experience feel polished is also what makes it difficult for screen reader users to navigate.
notable bug fixes
A few fixes from week 29 that are worth knowing about if you've hit them:
A startup hang that occurred when the Claude in Chrome extension was installed but Chrome wasn't running is fixed. A 300ms delay before async content appeared in Settings tabs, Stats, and diff views is fixed. A bug where /loop would hide a session from /resume after a single use is fixed. Background jobs running on LLM gateway auth no longer fail with "Not logged in" when the daemon respawns them.