Claude Code 2.1.116

💡

Up to 67% faster `/resume` on large sessions, quicker MCP startup, inline thinking progress indicator, smoother full-screen terminal scrolling, and a sandbox dangerous-path safety fix.

#Claude Code#Update
🔗 Official announcement →

This article is a summary based on official documentation.

What changed

Claude Code 2.1.116 (April 20, 2026) — /resume on large sessions is dramatically faster, MCP startup is quicker with multiple stdio servers, the thinking spinner shows progress inline, VS Code / Cursor / Windsurf full-screen scrolling is smoother, and a sandbox auto-allow that was bypassing dangerous-path checks is now blocked.

New features & improvements

  • Faster /resume on large sessions — up to 67% faster on 40MB+ sessions; handles sessions with many dead-fork entries more efficiently.

  • Faster MCP startup — resources/templates/list is deferred to first @-mention instead of preloading at startup, so multiple stdio MCP servers start faster.

  • Smoother full-screen scrolling — /terminal-setup now configures the editor’s scroll sensitivity for VS Code, Cursor, and Windsurf integrated terminals.

  • Inline thinking progress — the thinking spinner shows progress inline (“still thinking”, “thinking more”, “almost done thinking”), replacing the separate hint row.

  • /config value search — searching matches option values too; e.g. “vim” finds the Editor mode setting.

  • /doctor mid-response — can now be opened while Claude is responding, without waiting for the current turn to finish.

  • Auto-install missing plugin deps — /reload-plugins and background auto-update install missing plugin dependencies from marketplaces.

  • gh rate-limit hint — the Bash tool surfaces a hint when gh hits GitHub’s API rate limit, so agents can back off instead of retrying.

  • Usage tab — the Settings Usage tab shows 5-hour and weekly usage immediately and no longer fails when the usage endpoint is rate-limited.

  • Agent hooks: on main thread — agent frontmatter hooks: now fire when run as a main-thread agent via --agent.

  • “No commands match” — slash command menu shows this when the filter has zero results, instead of disappearing.

Security

  • Sandbox auto-allow no longer bypasses dangerous-path checks — rm / rmdir targeting /, $HOME, or other critical system directories still require confirmation even under sandbox auto-allow.

Notable fixes

Terminal rendering

  • Devanagari & other Indic scripts — broken column alignment fixed.
  • Ctrl+- undo on Kitty protocol — works in iTerm2, Ghostty, kitty, WezTerm, Windows Terminal.
  • Cmd+Left/Right on Kitty protocol — jumps to line start/end correctly.
  • Ctrl+Z hang via wrappers — no longer hangs the terminal when launched via npx, bun run, etc.
  • Scrollback duplication in inline mode — resizing the terminal or large output bursts no longer repeats earlier history.
  • Modal search dialogs — no longer overflow the screen at short terminal heights.
  • VS Code integrated terminal — scattered blank cells and disappearing composer chrome during scrolling fixed.

Sessions & commands

  • Cache control TTL 400s — intermittent API 400 when a parallel request completed during request setup; fixed.
  • /branch 50MB limit — no longer rejects transcripts larger than 50MB.
  • /resume silent failure — no longer silently shows an empty conversation on large session files; reports the load error instead.
  • /plugin duplicate items — the Installed tab no longer shows the same item twice when it also appears under Needs attention or Favorites.
  • /update and /tui after worktree — both work after entering a worktree mid-session.

Notes

  • If you work with long-running sessions, the /resume improvement is the headline change — especially on 40MB+ files.
  • Environments with multiple stdio MCP servers should see noticeably faster startup; worth re-baselining your cold-start profile.
  • Sandbox auto-allow is back to enforcing the dangerous-path safety check — automation scripts that relied on the bypass will see prompts again for critical paths.