Claude Code Gets Push Notifications, Fullscreen TUI, and Smarter Session Resume

Across versions 2.1.110 to 2.1.126, Claude Code shipped five updates that each solve a specific friction point in how developers actually work. None of them are headline-grabbing model improvements, but if you use Claude Code regularly, at least two of these will probably change your daily workflow.

Here is what landed, and what it means for you.

Mobile Push Notifications When Remote Control Is Active

Remote Control lets you monitor and interact with a running Claude Code session from your phone or browser, while the session itself continues running on your local machine. What it lacked was any way for Claude to reach out to you unprompted.

That changes with v2.1.110. With Remote Control active and “Push when Claude decides” turned on in /config, Claude will now send a push notification to your phone when a long-running task finishes or when it needs a decision from you to continue.

To set it up: Install the Claude iOS or Android app, sign in with the same account you use in the terminal, accept the notification permission prompt, then enable the setting in /config. If you toggle it on but notifications don’t arrive, open the Claude app on your phone first so it can register its push token. The /config screen will show “No mobile registered” until that happens.

For anyone running agentic tasks that take five, ten, or thirty minutes, this removes the need to babysit your terminal. Start a task, walk away, and your phone tells you when Claude is done or stuck.

One practical note: iOS Focus modes and Android battery optimisation can both delay or suppress notifications. If you are not seeing them reliably, check your OS notification settings and exempt the Claude app from battery optimisation on Android.

/tui Fullscreen Rendering Mode

Also in v2.1.110, Claude Code gained a /tui command and a --tui launch flag. Running /tui fullscreen switches the current session to a fullscreen, flicker-free rendering mode without starting a new conversation.

If you spend significant time in Claude Code inside tmux, iTerm2 split panes, or any terminal multiplexer, the standard output mode can feel visually noisy as long responses stream in. TUI mode eliminates the flicker and gives you a cleaner, more stable view.

You can set tui in your config file to make fullscreen the default on every launch. There is also a new autoScrollEnabled config option if you prefer to control scrolling manually rather than having the view follow output automatically.

Keyboard shortcuts were reorganised alongside this: Ctrl+O now toggles between normal and verbose transcript modes, and the new /focus command handles focus view toggling separately from the TUI toggle.

PR URL Paste in /resume

This one is small but genuinely useful. From v2.1.122, you can paste a pull request URL directly into the /resume search box, and Claude Code will find and surface the session that originally created that PR.

It works across GitHub, GitHub Enterprise, GitLab (including self-hosted instances), and Bitbucket. The first character of a paste triggers search mode automatically, so the workflow is: type /resume, paste the PR URL, press Enter.

Previously, if you wanted to return to the session behind a PR, you were searching by date or branch name and hoping your memory was good. Now the PR itself is the lookup key.

ANTHROPIC_BEDROCK_SERVICE_TIER Environment Variable

For teams running Claude Code through Amazon Bedrock, v2.1.122 added a new environment variable: ANTHROPIC_BEDROCK_SERVICE_TIER.

Set it to default, flex, or priority, and the CLI automatically forwards the value as the X-Amzn-Bedrock-Service-Tier header on every request. No SDK changes, no JSON config edits required.

flex offers a roughly 50% cost reduction on agent calls. priority targets faster response times. The practical tradeoff: Flex requests can be downgraded if the flex pool for that model is under load. AWS returns the tier it actually served you in ResolvedServiceTier, so you can verify what you got.

For enterprise teams running large volumes of agentic Claude Code sessions through Bedrock, this is a cost control lever you can flip at the environment level across all machines without touching any application code.

claude project purge

Added in v2.1.126, claude project purge [path] deletes all Claude Code state for a given project: transcripts, tasks, file history, and the project’s config entry.

A busy project can accumulate substantial transcript JSON over time, and there was previously no clean way to wipe it short of manually hunting down the state directory. The new command handles that in one step.

The command supports several flags worth knowing:

• --dry-run shows what would be deleted without deleting anything
• -y / --yes skips the confirmation prompt
• -i / --interactive lets you choose which categories of state to remove, rather than deleting everything
• --all purges state across every project Claude Code has tracked

Beyond storage, there is a privacy consideration here. Transcripts can contain API keys, environment variables, internal business logic, and other sensitive content that accumulates naturally during a coding session. Before handing off a machine or sharing a project directory, claude project purge gives you a clear, auditable way to remove that history.

It is worth distinguishing this from /compact and /clear, which manage the active session window in memory. project purge targets persisted disk state, which those commands leave untouched.

---

These five updates do not change what Claude Code can do at the model level, but they close real gaps in how it fits into a working day: staying informed while away from the terminal, cleaner visual output for extended sessions, faster navigation back to past work, cost control for cloud deployments, and clean handoffs between developers or machines.

All five features are available now. If you are not on v2.1.126 yet, npm update -g @anthropic-ai/claude-code will get you there.