Claude Code Can Now Publish Live Pages, Test iOS Apps — and It Finally Runs on Linux

Anthropic has rolled out three additions to Claude Code that change what the coding agent produces and where it runs. It can now publish a session as a live, shareable web page — one that fetches fresh data for each viewer. It can build an iOS app, launch the simulator and check the result without leaving the session. And it finally has a desktop app for Linux.

Now Claude Code can turn a session into a live page

Artifacts take the work from a session and publish it as an interactive web page. Anthropic describes them as "live, shareable visual pages" that "update themselves as your session works" — when Claude modifies an artifact, the open page refreshes in place and anyone viewing sees the change immediately. Each publish creates a new version at the same link, with version history.

Use it for	Instead of
PR walkthroughs and code reviews	A wall of terminal output pasted into chat
Architecture maps and system explainers	A diagram that goes stale the day it's drawn
Dashboards with filtering and sorting	A spreadsheet someone has to regenerate
Incident timelines and release checklists	A doc updated by hand during the incident
License audits, data-flow maps, security findings	A one-off report nobody reruns

Every artifact is private to its author by default. That's worth flagging because it changed: at launch in June, artifacts were viewable only by authenticated members of your organisation and explicitly could not be made public. They can now optionally be shared through a public link.

Now those pages can fetch live data

The bigger shift is that artifacts can call MCP connectors at view time. The page isn't a snapshot — it queries a connected system when someone opens it.

The credential model is the clever part: each viewer authenticates with their own Claude account and hits their own instance of the artifact using their own connections. No API keys are embedded in the page, and there's no cost to whoever created it. A dashboard you publish shows each teammate the data they're entitled to see, not the data you can see.

There's a deliberate boundary: connector-backed artifacts work on Pro, Max, Team and Enterprise plans, and are not available on publicly-shared artifacts. A page anyone on the internet can open cannot reach into a connected system — which is the right call, and explains why public sharing and live data are mutually exclusive.

Taken together, this moves the output of a coding agent from "here's a script" to something closer to "here's a small internal tool, deployed." That is a meaningfully different product than a terminal that writes files.

Now Claude Code can build and test an iOS app itself

In beta, Claude Code can compile your iOS app, launch the simulator and verify the result without you leaving the session — closing the loop that previously required a human to switch to Xcode, hit run and report back what happened.

This is the pattern that matters most for agentic coding: an agent that can only write code has to be told whether the code worked, and an agent that can run and observe its own output can iterate on its own. The constraint is platform-bound — it's in the desktop app, macOS only, and requires Xcode.

And Claude Code is available on Linux

The Claude desktop app is now in beta for Ubuntu and Debian, distributed through its own apt repository so updates arrive with normal system updates. It carries the same Chat, Cowork and Code sections as the macOS and Windows builds.

On the Linux desktop app	Status
Parallel sessions with Git isolation	Yes
CI monitoring with auto-fix and auto-merge	Yes
Inline diff review	Yes
Your CLAUDE.md, MCP servers and skills	Carry over from the CLI
Computer Use	Not in the Linux beta
Voice dictation	Not in the Linux beta

The config detail is the practical one: the desktop app shares configuration with the CLI, so the CLAUDE.md, MCP servers and permission settings you already have keep working. Linux has been the conspicuous gap for a tool whose core users overwhelmingly deploy to Linux servers, and closing it removes a real reason developers stayed on the terminal-only CLI.