All posts
Production Published 12 min

Claude Code Week 29: live MCP artifacts change the blast radius of a shared dashboard

Claude Code Week 29 (July 13–17, v2.1.207–v2.1.212) lets published artifacts call MCP connectors on each view through the viewer's own connections, with first-call approval. Public sharing links, editor roles, screen reader mode, and auto-background for long MCP calls shipped in the same window. Treat live artifacts like production integrations, not pretty exports.

Jigar JoshiJigar JoshiAgentic AI Architect and Consultant
In this post (5 sections)

Introduction

In June, Artifacts were mostly review captures: org-private pages built from session context. I covered that shape in Claude Code Artifacts live review pages. Week 29 changes the contract. A dashboard can pull open PRs through GitHub on every load. That is useful. It is also a standing OAuth and data-exfiltration surface if you treat "Share" like sending a screenshot.

Pair this with Week 28's desktop browser and /doctor checklist. Browser + live connectors + public links is a different risk class than a CLI coding session.

What Week 29 changes

Week 29 features that need an org decision
FeatureWhat it doesDecision
Artifacts call MCP connectorsLive reads/actions on view via viewer's connections; first-call approvalAllowlist which connectors artifacts may invoke; ban write tools on shared pages
Public sharing linksArtifacts shareable beyond org-only review pagesDefault private; public requires security review
Editor roles (Team/Enterprise)Shared editing on artifactsLimit who can edit live connector-backed pages
Screen reader modeLinear text UI for VoiceOver/NVDA (--ax-screen-reader / CLAUDE_AX_SCREEN_READER)Enable in accessibility standards; test permission prompts
Long MCP auto-backgroundCalls >2 min background automaticallyTune CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS; monitor orphaned jobs
Session capsWebSearch and subagent spawns default cap 200Keep caps; lower for contractor seats

Live artifact governance checklist

  1. 01
    Separate read dashboards from action dashboards
    A page that lists open PRs is not the same as a page that merges them. If an artifact can take actions, require the same gates you use for unattended automation in governing agent autonomy.
  2. 02
    Force viewer-scoped connectors, never shared bot tokens
    Week 29 runs calls through the viewing account. Do not "fix" demos by wiring a shared service account into every viewer's Claude settings. That recreates the blast radius EMA was meant to shrink. See MCP EMA.
  3. 03
    Review public link policy before enabling
    Public sharing plus live MCP is closer to publishing an internal app than posting a design mock. Default off for regulated teams. Log who created public links.
  4. 04
    Ban customer-PII connectors on shareable artifacts
    Gmail, CRM, and ticket systems do not belong behind a link that can forward. Same instinct as Cowork connector allowlists in the Cowork playbook.
  5. 05
    Watch long MCP calls after auto-background
    Two-minute auto-background keeps the session usable and can hide runaway tool spend. Wire duration and cost into the agent observability stack.
  6. 06
    Pilot screen reader mode on permission flows
    Accessibility is not optional. Have someone who uses VoiceOver or NVDA approve a permission prompt and complete a short task before you claim Week 29 is rolled out.

Other Week 29 defaults worth locking

  • "Always allow" rules now save at repository root across worktrees. Audit what engineers permanently allowed.
  • Auto mode no longer needs CLAUDE_CODE_ENABLE_AUTO_MODE on Bedrock, Google Agent Platform, and Microsoft Foundry; admins can disableAutoMode. Confirm cloud defaults match policy.
  • Amazon Bedrock, Google Agent Platform, and Claude Platform on AWS defaulting to Opus 4.8 changes spend. Update the unified spend dashboard before finance notices.

Conclusion

Week 29 makes artifacts honest about being apps. Live MCP on view is the feature. Viewer-scoped OAuth, link policy, and spend caps are the work. Ship the version bump only after those three have owners.

Sources: Claude Code Week 29 changelog at Claude blog (releases v2.1.207–v2.1.212).

The weekly take

Agentic AI patterns, delivered Thursdays

What I am shipping, watching, and pruning out of client stacks each week. One email. No fluff.

Shipping an agentic AI project this quarter?
Book a 30-min consult
Frequently asked

Questions readers ask about this post

Share this post
LinkedIn Facebook

Read next

Production

Gemini 3.6 Flash shipped, 3.5 Pro did not: the routing checklist I run after July 21

On July 21, 2026 Google released Gemini 3.6 Flash ($1.50/$7.50 per million), 3.5 Flash-Lite ($0.30/$2.50, ~350 tok/s), and a limited-access 3.5 Flash Cyber via CodeMender. Gemini 3.5 Pro stays in partner testing. Here is how I re-benchmark cost per completed task against Sonnet 5 and Fable without betting the roadmap on Pro.

Read post Production

Claude Code Week 28: the desktop browser and /doctor checklist I run before enabling it org-wide

Claude Code Week 28 (July 6–10, releases v2.1.202 through v2.1.206) shipped a sandboxed in-app browser on Desktop and upgraded /doctor from a read-only report into a repair tool. Auto mode also blocks transcript tampering. Here is the governance checklist I run before teams turn browsing and auto-fix on for every engineer.

Read post Production

Claude Cowork is not Claude Code for civilians: the knowledge-worker playbook after the mobile launch

Anthropic shipped Claude Cowork on mobile and web July 8, starting with Max subscribers. Usage data from 1.2 million sessions shows more than 90% of Cowork work is non-technical: memos, RFP reviews, inbox triage, decks. Tasks run in the cloud, sync across devices, and continue when you close the app. Here is how I govern Cowork without treating it like a coding agent.

Read post