Claude Code v2.1.132: SIGINT Graceful Shutdown, MCP Fixes, and Terminal Handling Overhaul

Anthropic shipped Claude Code v2.1.132 with a heavy focus on terminal reliability, MCP server robustness, and edge-case fixes that matter when you're running AI coding agents for hours on end. Here's what changed for developers and CI pipelines.
Environment Variables and Session Plumbing
CLAUDE_CODE_SESSION_IDis now exported into the Bash tool subprocess environment, matching thesession_idpassed to hooks. Useful for tracking tool calls back to a session.CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN=1lets you opt out of the fullscreen alternate-screen renderer, keeping the conversation in the terminal's native scrollback.
Graceful Shutdown and Resume Fixes
- External
SIGINT(e.g., IDE stop button,kill -INT) now triggers a graceful shutdown: terminal modes are restored and the--resumehint is printed instead of an abrupt exit. --resumeno longer fails withno low surrogate in stringwhen a tool error truncation split an emoji; pre-corrupted sessions are sanitized on load.--permission-modeflag is now respected when resuming a plan-mode session with-p --continue / --resume, and plan mode is re-applied afterExitPlanModewithin the same session.
Terminal and Input Handling
- Fullscreen mode no longer shows a blank screen after laptop sleep/wake or Ctrl+Z/
fguntil the next keystroke or stream output. - Cursor now lands correctly on Ctrl+E/A/K/U/arrow keys when an Indic conjunct or ZWJ emoji wraps across lines.
- Vim operators no longer corrupt text containing decomposed (NFD) accented characters.
- Pasting text starting with
/no longer swallows input or triggers an unknown-command reply. - Stray escape sequences from focus events or mouse-tracking reports interleaved with bracketed paste are no longer dumped into the prompt.
- Mouse wheel scrolling speed normalized in Cursor and VS Code 1.92–1.104 (upstream xterm.js bug) and in JetBrains IDE 2025.2 terminals (spurious arrow keys, wrong-direction events, runaway acceleration).
/usageCtrl+S no longer hangs when copying the stats screenshot to the clipboard on Linux/X11./terminal-setupno longer shows a contradictory error in Windows Terminal — Shift+Enter is natively supported there./effortpicker now reflects theCLAUDE_CODE_EFFORT_LEVELenv var override./statusfixed for showing the wrong default model for some users.- Slash command autocomplete popup now scales vertically with terminal height instead of being capped at ~3–5 visible commands.
- Statusline
context_windowtoken counts now reflect current context usage, not cumulative session totals. - Alt+T (thinking toggle) works on macOS terminals without 'Option as Meta' enabled (iTerm2, Terminal.app defaults).
- Dead keyboard input on Windows after re-opening a background session from
claude agentsis fixed.
MCP Server Stability
- Unbounded memory growth (10GB+ RSS) when a stdio MCP server writes non-protocol data to stdout is fixed.
- MCP servers that connect but fail
tools/listnow retry once and show 'connected · tools fetch failed' in/mcpinstead of silently showing 0 tools. - Unauthorized claude.ai MCP connectors now show as 'needs auth' instead of 'failed', and headless
-pmode no longer retries non-transient 4xx connection failures.
Other Improvements
- Visual consistency improved in slash command dialogs and
/login,/upgrade,/extra-usagedialog spacing. /tui fullscreenstartup banner updated to describe additional renderer benefits (lower memory usage, mouse support, auto-copy on select).- Fixed Bedrock and Vertex 400 errors when
ENABLE_PROMPT_CACHING_1His set.
This release is particularly relevant for developers running Claude Code in headless CI environments (where graceful shutdown and session resume matter), users of IDE terminals with non-standard escape handling, and teams relying on MCP servers for custom tool integrations.
📖 Read the full source: GitHub Claude-Code
👀 See Also

AI Subscription Pricing Crash: Why Your Enterprise Bill Is About to 10x
AI labs like OpenAI, Anthropic, and Microsoft are losing money on every subscription seat. Agentic workloads have broken the flat-fee model — GitHub Copilot moves to usage-based billing June 1, 2026. Enterprises that built on subsidized pricing face a correction.

Gemma 4 Chat Template Bug: Tool Parameters with anyOf/null Rendered as Empty type
A bug in Gemma 4's chat template drops $ref, anyOf, and $defs from tool parameter schemas, rendering nullable refs as empty type fields. A Jinja fix restores correct schema parsing for all inference engines.

EU Forces Google to Open Android AI to Third Parties Under DMA
European Commission proposes measures to allow third-party AI assistants system-level access on Android, including hot word invocation, screen context, and local model hardware access. Google calls it 'unwarranted intervention'.

Research Findings on AI Agent Reliability and Development Patterns
A collaborative research session with Claude Opus analyzed 15 papers on AI agents, revealing quantified reliability problems: agents produce 2-4 different action sequences across 10 runs, with 69% of divergence occurring at the first decision. Self-improving agents showed safety refusal rates dropping from 99.4% to 54.4% through their own learning.