Claude artifacts can call viewers' own MCP connectors

Published Claude artifacts can invoke each viewer's own MCP connectors, per Week 29's release notes from code.claude.com.

That's a real shift for anyone who's tried to share an internal dashboard before. A dashboard built once can render live data and take actions for whoever opens it, instead of freezing the state from whoever built it. Each call runs through the viewing account's own connections, and viewers approve access before the first call goes through.

I've watched every attempt at a shareable tool hit the same wall: bake in one person's API keys, or make everyone run their own copy. Resolving connectors per viewer instead of per session closes that gap. The artifact carries the logic. The account viewing it supplies the access.

Week 29 also shipped public sharing links and editor roles on Team and Enterprise accounts. It let people publish artifacts straight from Claude Tag sessions too, per the release notes. Together, that's enough to replace a chunk of the small internal tools teams build and maintain forever.

The release notes don't say whether an admin can see which connectors a published artifact is calling, or revoke a viewer's approval after granting it. For anything shared inside a company, that's the detail that decides whether this gets used for real work or stays a demo.

This turns artifacts into a real internal-tool category. I've built enough throwaway ops dashboards to know why that matters. Teams will start wiring a published artifact to a connector instead of standing up a framework app for something read-only. A lot of those quick versions will end up staying the real thing.