Anthropic's agentic command-line coding tool. Updates to the Claude Code CLI, IDE integrations, and SDK.
Major update adding file mention environment variables, idle token compaction for prompt caching, new APIs for mods and plugins (including image attachment handling and session identity), improved agent and skill controls, better tool management for eval runs, and numerous behavioral changes to plugin hooks, MCP requests, and session handling.
This release adds new configuration options for managed policies, subagent behavior, workflow models, and retry handling. It includes improvements to the Claude apps gateway, Desktop Code tab settings, file reading capabilities, and session management across cloud and self-hosted environments. Pricing for Sonnet 5.5 cache reads has been updated to $0.10 per million tokens, and various UI and behavior refinements have been made to Claude Tag, VSCode, Chrome, and Slack integrations.
This release adds onFailure blocking for hooks, Program Status Protocol support for terminal status display, upstream timeout configuration, improved tool descriptions, and numerous UX enhancements including file handling in chat, better error messages, and protocol improvements across VSCode, Claude Tag, and MCP integrations.
Claude Code 2.1.287 introduces Claude Mods for deeper plugin behavior customization, adds the "You should know" built-in mod for monitoring, and brings URL prompts from MCP servers, improved filtering in the agents view, and numerous UX enhancements. The update also includes support for 1M context windows, background shell execution in VSCode, and security improvements for file operations.
This release adds support for newer Claude Desktop keys in policy blocks, assumes IAM roles on Bedrock upstreams, introduces guardrail support, adds MCP URL-mode elicitation, and includes a fast mode switch for web sessions. It also changes how auto mode handles classifier reviews, improves session artifact handling, and refines the UI for various lists and dialogs across VSCode, JetBrains, and web platforms.
Fixed a bug where prompt and agent hooks written as blocking instructions were not working correctly. Improved evaluation of Stop and SubagentStop hooks so Claude is less likely to stop early when hooks are written as conditional instructions.
Claude Haiku 5.5 is now the default Haiku model on the Anthropic API with 1M context and updated pricing. Added agentType to subagent payloads for type identification and isDeferred parameter for tool schema visibility control. Reverted previous auto mode denial behavior and cloud session sleep fixes, adjusted skill syncing intervals, and improved logging performance for MCP resources and agents.
Released version 2.1.293 with Claude Haiku 5.5 as the default Haiku model featuring 1M context window, expanded agent functionality with agentType and isDeferred fields, and numerous bug fixes across context management, MCP connections, vim mode, permissions, and cloud sessions.
Claude Code 2.1.292 includes new features like marketplace source support for plugin installation, effort parameters for agent tools, and prompt caching for mods. This release also addresses numerous security fixes including permission bypass issues, file access restrictions, and cache tampering vulnerabilities, plus improvements to session handling, auto mode restoration, and various UI/UX fixes.
This release adds marketplace plugin installation via --marketplace flag, effort parameters for sub-agents, prompt caching for mods, workflow agent spawning, and improved analytics in Code Review. It also includes various improvements to plugin testing, usage messaging, artifact handling, and administrative interfaces across Claude Tag and other features.
Fixed a regression where cloud sessions could drop answers to permission prompts and another regression where the last messages of a session could be lost when quitting.
This release adds new capabilities for plugin hooks including serverToolUses, agentId, and ceiling fields, introduces Deny button for sign-in approval, adds managed agents setup commands, and includes numerous improvements to permissions, logging, WebSearch budgeting, and IDE integrations across VSCode, Chrome, and Slack.
This release adds new plugin hook capabilities for serverToolUses and agentId tracking, introduces Managed Agents CLI commands, improves session handling and task scheduling, and fixes numerous issues with proxy compatibility, long sessions, caching, URL patterns, and plugin loading behavior.
This release fixes multiple security vulnerabilities including unauthorized approval bypass for managed mods, plugin permission bypasses, and MCP server description rewriting. It also addresses terminal freezes with malformed script tags, file access denial rule bypasses via symlinks, authentication regression, and numerous plugin/mod rendering and session stability issues.
Added agent.spawn for teammates with one agent ID across plugin hook events, plus idle and waiting states in $.agent.list(). Improved large file opening performance in plugin code panes and reverted a VSCode authentication change that increased sign-outs. Enhanced error messages for failed mod bands or panes.
Added text selection API for fullscreen mode, built-in GitHub API for cloud sessions, Ctrl+C prompt recovery, MCP OAuth re-authentication, code review finding limits control, and session search shortcuts. Improved background command time limits, auto mode classifier, and MCP URL prompt behavior.
This release adds new UI features like selection tracking and session search, improves conversation resilience with better recovery and auto-compaction, fixes numerous issues with MCP servers, plugins, and cloud sessions, and hardens permission prompts and authentication flows.
Claude Code 2.1.287 introduces Claude Mods for plugin-based behavior modification, adds the "You should know" built-in mod for oversight, improves the agents view with session filtering, and includes extensive fixes for remote sessions, accessibility features, tool heartbeats, model management, permission handling, and various edge cases across multiple platforms.
This release includes numerous bug fixes and improvements across permissions UI, authentication, session management, MCP servers, remote control, and message handling. Key fixes address API errors in resumed sessions, cloud session loading issues, credential handling on macOS, and various edge cases with subagents, plugins, and file attachments.
Release 2.1.286 introduces new features including permission prompt counters, mouse support for list navigation, VSCode bookmarks panel, conversation question tracking, and channel management in Claude Tag admin settings. Multiple improvements enhance API retry logic, MCP server connections, list layout consistency, and UI refinements across the platform.
Release 2.1.285 adds environment variables for controlling WebFetch and timeout retries, new CLI commands for desktop and plugin configuration, support for configuring MCP servers at install time, and the allowedProviders managed setting to restrict API providers. The release also includes numerous bug fixes for SSH plugin installs, artifact handling, model switching, MCP server management, cloud sessions, and various stability and security improvements.
Claude Code 2.1.285 adds new environment variables and CLI commands for controlling web fetch and MCP server configuration, introduces plugin management enhancements, adds allowedProviders settings for API provider control, and includes numerous changes to MCP tools, background command handling, code review workflows, and various platform-specific improvements.
Claude Code 2.1.284 introduces Claude Sonnet 5.5 as the default Sonnet model with 1M context and improved pricing, adds flexible permission prompting with "ask again next time" option, and includes numerous UI improvements like message timestamps, better MCP server management, and the Ultracode toggle. The release also brings security enhancements including certificate-based client authentication and improved retry behavior.
Claude Code 2.1.284 adds Claude Sonnet 5.5 as the default model with 1M context window and improved pricing, introduces new UI controls for effort levels and rate limits, adds MCP server reconnection, and fixes numerous bugs including response stream corruption, session model unavailability handling, MCP tool failures, and various terminal rendering and interaction issues.
Version 2.1.283 adds new features including gateway hint headers for request grouping, model availability management settings, expanded observability with MCP tool outputs in OpenTelemetry spans, and a /doctor prompt-audit command. It also includes numerous bug fixes across SDK sessions, MCP servers, plugin validation and loading, keybindings, Git operations, and remote control functionality.
Claude Code 2.1.283 introduces gateway hint headers for LLM request grouping, model management settings for availability control and blocking, expanded OpenTelemetry logging for tool outputs, prompt audit capabilities, and UI improvements including click-to-expand messages and Slack integration enhancements. Updates also include performance optimizations for artifact watching, git lifecycle hook improvements, and various behavioral refinements across terminal and cloud session features.
Added maxProseWidth setting for prose in wide terminals, startup telemetry notices, Chrome managed settings, database failover resilience, and cloud session repository features. Updated API billing docs links, auto mode classifier behavior, and settings precedence rules; refined UI styling and Slack card rendering.
Includes new prose width setting, telemetry reporting, managed MCP support, and numerous fixes for extended thinking, session resumption, permissions, rendering, and third-party provider integrations across API, terminal UI, and plugin systems.
This release adds Claude apps gateway support for newer desktop policies, Bedrock assume_role and guardrail features, telemetry enhancements, and MCP URL-mode elicitation. It also includes numerous fixes for session resumption, prompt cache handling, stream processing, permission dialogs, and sandbox behavior to improve stability and correctness across interactive and headless sessions.
Release 2.1.281 adds Claude apps gateway support for desktop policy blocks, guardrails, and telemetry improvements; includes MCP URL-mode elicitation, validation enhancements, and a comprehensive set of fixes addressing session resumption, prompt caching, streaming errors, tool calls, permission handling, and sandbox functionality.
Released Claude Code 2.1.280 with Claude Opus 5.5 support, mouse controls for fullscreen mode, MCP configuration options, and numerous fixes to auto mode, symlink handling, dialog interactions, text input, conversation validation, session state management, and background agent communication.
Auto mode now defaults to server-side classifier for Claude API, Enterprise, and Bedrock/Vertex/Foundry/gateway users, eliminating classifier overhead charges. A new Auto mode server row in /status shows whether the session's classifier runs server-side.
2.1.277 includes support for AGENTS.md project instructions, enhanced Claude apps gateway configuration options for proxies and headers, numerous bug fixes for hanging sessions and error reporting, corrections to conversation history handling, and improvements to plugin management, tool reliability, and UI stability.
Claude Code 2.1.277 adds AGENTS.md support as a fallback to CLAUDE.md, introduces new gateway proxy configuration options, and fixes numerous bugs affecting sessions, tool execution, plugin management, and UI stability across various scenarios.
Fixed a regression from 2.1.275 where all requests would fail with a 400 error when ANTHROPIC_BASE_URL is configured to point at a proxy or gateway. The issue was related to the 'advisor_20260301' input tag handling.
Added signed-in account confirmation in gateway sign-in, a send-now key for interrupting turns, startup warnings for telemetry export failures, and syncing of Claude.ai skills and plugins to terminal sessions. Fixed numerous issues including memory file caching, subagent message handling, file suggestion ordering, fullscreen mode display bugs, plugin marketplace updates, credential leakage in logs, vim mode cursor placement, file diff performance, malformed transcript handling, permission rule generation, and improved prompt caching and artifact tools.
This release adds signed-in account confirmation in the gateway, a send-now keyboard shortcut (Ctrl+Enter), skill/plugin syncing from claude.ai accounts, and plugin marketplace installation. It includes numerous bug fixes for memory file caching, file suggestions, fullscreen mode, plugin URLs, session resumption, and vim mode, plus improvements to prompt caching, artifact publishing, and image handling.
Claude Code 2.1.274 adds critical memory usage warnings, OpenTelemetry instrumentation improvements, MCP server connectivity enhancements, and fixes numerous bugs across session management, transcript handling, tool execution, and shell integration to improve stability and user experience.
This release adds new request headers for LLM gateways, improves MCP server handling with disconnection notifications and automatic reconnection, adds session forking from Remote Control, and fixes numerous issues including permission checking, settings handling, authentication errors, context management, and artifact publishing.
This release includes bug fixes and reliability improvements to enhance overall stability and performance.
This release adds fast mode for Remote sessions, mouse support in fullscreen config panels, per-command domain sandboxing for security, and numerous fixes for policy caching, MCP server handling, Bash permission checks, and session resumption behavior.
Resolved a regression where read-only git commands in Bash would unexpectedly prompt for permission after a session had been running for some time. This issue was introduced in version 2.1.269.
Release 2.1.269 includes a new claude plugin eval command for testing plugins, output style switching, workflow concurrency improvements, and numerous bug fixes addressing prompt cache issues, terminal handling, permissions, and session management across remote and headless deployments.
Extensive release including gateway pricing configuration, improved security warnings, new managed settings for administrators, expanded CLI JSON output support, artifact icons, and numerous bug fixes across MCP tools, WebFetch, permission handling, model caching, session management, and UI elements.
This release adds maxEffortLevel settings to cap effort across providers, introduces a --system-prompt-snapshot option for iterating on prompts, and fixes numerous stability issues including scheduled tasks failing at startup, rendering problems on mobile, session resumption bugs, MCP tool handling, and prompt-cache reuse errors.
Fixed a regression in 2.1.265 where the CLAUDE_CODE_USE_GATEWAY environment variable was incorrectly forcing Cloud-gateway sign-in even when used with API keys or custom auth headers, causing request failures. The variable is now ignored when used alone, restoring previous behavior without requiring configuration changes.
This release includes numerous bug fixes for prompt-cache reuse, tool result handling, session resumption, and remote control functionality, along with improvements to plugin management, keyboard shortcuts, and performance for large repositories and long sessions.
This release includes bug fixes and reliability improvements to enhance overall stability and performance.
A comprehensive update adding organization policy status visibility, expanded output buffer settings, improved subagent system prompt handling, skill diagnostics, and numerous bug fixes for input handling, cloud sessions, remote control functionality, and proxy compatibility. Many stability and performance improvements across SSH, plugin syncing, and concurrent operations.
This release adds a diff panel to show uncommitted changes, new debugging features like /reload-plugins and /advisor text form, support for newer Claude Desktop keys, and fixes numerous issues with permissions, model switching, prompt caching, file handling, and agent teams.
This release adds managed MCP server settings for organizations, improves permission handling with unattended mode, enhances GitLab merge request display, and fixes numerous bugs including concurrent session conflicts, caching issues, file permission handling, and remote session stability.
Claude Code no longer fails to launch on macOS 12 Monterey, fixing a regression from version 2.1.255. Additionally, remote and scheduled sessions no longer fail with content validation errors when permission approvals are re-sent.
This release introduces Claude Fable 5.1 as the default model with 1M context and improved pricing, adds time format and timezone settings, enhances security with a Containment Escape rule, and fixes numerous issues including session management, cache behavior, sandbox handling, and background process reliability across desktop and remote control scenarios.
Fixed Bash command failures related to task output swap on some Macs, corrected "always allow" settings not persisting in projects without .claude/settings.local.json, resolved Remote Control session stalling after tool completion with degraded connections, and fixed background task notifications with large failure output exceeding API limits.
This release adds hook events for model switching, live streaming of subagent tool calls to Remote Control clients, spend limit tracking, and per-session prompt-cache metrics. It also includes numerous bug fixes for file operations, security issues, symlink handling, session management, and various CLI and UI improvements.
This release includes bug fixes and improvements to system reliability and stability.
Claude Code 2.1.248 adds restricted mode to disable built-in tools, experimental prompt cache TTL configuration, server-managed settings diagnostics, and cross-session messaging. The release includes numerous fixes for prompt cache misses, session management, permission handling, MCP servers, and terminal interaction issues.
This release adds the SendFeedback tool for drafting error reports, enhances API cost profiling via /claude-api cost-optimize, improves plugin marketplace hardening, and fixes numerous bugs affecting keyboard shortcuts, background sessions, shell commands, and terminal integration across various keyboard layouts and edge cases.
A comprehensive release featuring UI improvements including completion times and auto mode classifier rules, along with numerous stability fixes across fullscreen rendering, transcript performance, background sessions, markdown rendering, plugin handling, and API compatibility with third-party endpoints.
Claude Code 2.1.243 adds loops usage breakdown, model picker customization, prompt cache TTL settings, and keyless sign-in via Anthropic Console. It also includes numerous fixes for MCP server recovery, auto mode availability, session resuming, cross-session messaging, spellcheck, authentication, plugin dependencies, and various UI/terminal rendering issues.
Fixed a startup crash on Linux distributions shipping glibc 2.44, including Arch Linux, CachyOS, and Fedora Rawhide. Users on affected systems can now launch the application without errors.
This release includes bug fixes and reliability improvements to enhance stability and performance.
This release includes bug fixes and reliability improvements to enhance overall stability and performance.
This release includes cost estimation improvements for data-residency workspaces, fullscreen renderer support on additional cloud platforms, Python SDK migration tools for anthropic 1.x, cloud session plugin syncing, Alpine/musl compatibility fixes, and numerous bug fixes addressing streaming issues, proxy handling, UI rendering, MCP servers, session management, and keyboard input handling.
This release adds a keybindingFlavor setting for readline-style editing, plugin marketplace headersHelper support for HTTP headers, self-hosted runner shutdown and proxy authorization options, and includes numerous bug fixes for memory management, session handling, Remote Control reconnection, and terminal compatibility.
Fixed prompt caching functionality for sessions using an LLM gateway or custom base URL. Added a new built-in "Concise" output style that leads with results and skips preamble, available in the Output style settings.
This release adds environment variable configuration for default models, cross-session messaging with idle notifications, and improved sandbox rules on macOS. It includes numerous bug fixes for clipboard operations, fullscreen rendering, the model picker, message handling, and terminal integration, plus improvements to startup performance, auto mode, and screen reader support in VS Code.
This release adds optional spellcheck for the prompt input, fixes issues with prompt caching, nested markdown lists, input highlights, permission dialogs, slash commands, and various UI behaviors. It also improves memory and CPU usage for background sessions, enhances error messages, and fixes issues in Vim mode and cross-session messaging.
This release adds project-specific config directories, GitLab merge request badges, automatic session continuation at usage limits, and improved email handling. It includes numerous bug fixes and security improvements across file access, permissions, markdown rendering, shell commands, and Remote Control sessions.
This release adds GitLab merge request support, user identity forwarding for apps gateway, memory limits for Bash tools, WebFetch cache configuration, and numerous bug fixes including cloud session stability, MCP connection handling, permission prompt notifications, CPU usage on Linux, and Windows credential leak prevention.
Claude Code v2.1.232 enables subagent forking by default, adds cross-session messaging via @mentions, improves remote control stability, and includes multiple security fixes for permission bypasses and credential redaction. The release also adds GitLab plugin marketplace support and enhances the managed settings system.
Resolved an issue where MCP OAuth sign-in would fail with a redirect URI mismatch error when using pre-registered OAuth clients like Slack. This fix ensures proper OAuth handling for servers with pre-configured clients.
This release includes numerous enhancements and bug fixes across Remote Control sessions, streaming stability, plugin marketplace support, and IDE integration. Notable improvements include SSE keepalive pings to prevent idle timeouts, fixes for crashes in terminal output and file handling, better workflow optimization, and new VSCode features like session groups and resizable side panels.
Claude Code 2.1.228 includes multiple bug fixes for interactive sessions, Git integration on Windows, session management, and plugin caching, along with improvements to cross-session messaging, credential handling, and tool behavior. This release also hardens skills synced from claude.ai to prevent shadowing local commands and improves overall stability.
This release fixes feature flag evaluation for expired login sessions, corrects Bash command failures in GitHub-hosted runners, resolves /tui conversation rewinding issues, improves slash-command menu styling and selection, and reduces performance stalls in file-not-found suggestions and at-mention size checks.
This release includes bug fixes and reliability improvements to enhance stability and user experience.
Claude Code 2.1.225 adds gateway spend-limit support with detailed usage warnings, workspace trust prompts for agents, and fixes numerous issues including transient authentication failures in headless sessions, MCP OAuth server reliability on macOS, safety-filter retry behavior, cross-session message handling, conversation history corruption on Remote Control resume, session directory bugs, self-hosted runner initialization, web session reporting, Remote Control photo handling, Focus view folding issues, and improvements to SendMessage for starting conversations by name.
This release adds self-hosted environments for running Claude Code sessions on your own machines, introduces archive-based plugin installation, and adds cross-session messaging with improved sandbox credential masking. Multiple fixes address session directory resolution, message delivery reporting, filesystem deny entries, plugin installation corruption, and Remote Control reliability.
This release adds marketplace filtering via owner wildcards, improves permission and security controls including fixes for Bash permission bypasses and sandbox escapes, enhances model discovery and overrides handling, and refines session management and command review workflows.
This release fixes multiple security and stability issues including worktree isolation for git commands, tool restriction bypasses in background tasks, usage credit blocking, proxy connectivity, MCP server attribution, connector authorization display, and rare crashes. It also improves auto mode safety with permission classification for inter-session messages and refines diff viewing.
Claude Code 2.1.221 adds a Focus view to hide tool activity in VSCode, introduces credential file masking for sandboxes, and fixes numerous bugs across permissions, MCP server connections, plugin management, Vim mode, and session handling. Improvements include better tool search for Vertex AI, cache-efficient permission checks, and enhanced stats tracking.
This release includes various bug fixes and reliability improvements to enhance overall stability and performance.
Added Claude Opus 5 as the default model with 1M context and fast-mode support at $10/$50 per Mtok. Introduced sandbox network allowlist settings, DirectoryAdded hook for mid-session directory registration, improved MCP error handling and configuration validation, nested subagent forwarding in stream-json, and numerous bug fixes and UI improvements across the terminal, remote control, and model picker.
Claude Code now features Claude Opus 5 as the default Opus model with 1M context and fast mode pricing. This release includes numerous improvements to sandbox security settings, MCP server handling, workflow management, self-hosted runner reliability, UI/UX enhancements, and bug fixes for permission handling, model selection, and remote control functionality.
Claude Code 2.1.218 includes numerous fixes and improvements: /code-review now runs as a background subagent to avoid filling conversations, accessibility announcements were added for text deletions, Windows path corruption was fixed, and multi-line paste handling in terminals was improved. Many other fixes address issues with context reporting, tool execution, error messaging, screen reader support, MCP server connectivity, and session management.
This release refactors /code-review to run as a background subagent, fixes multiple critical bugs affecting file accessibility on Windows, terminal paste behavior, and accessibility features, and improves auto mode intelligence and trust dialog clarity.
This release adds emoji shortcode autocomplete in prompts, adds warnings for failing transcript writes and disabled session saving, and fixes multiple critical issues including a memory leak in MCP tool output truncation, Windows auto-update failures, background session directory canonicalization, auto-compact triggering, and various settings being ignored in Desktop sessions. Also fixes screen reader mode announcements, telemetry endpoint routing, transcript preview layout, subagent spawning controls, and budget enforcement for background agents.
This release adds sandbox filesystem control, fixes message normalization performance degradation in long sessions, resolves OAuth token rotation issues, and addresses numerous stability problems including file attachment failures, symlink handling, permission validation, and UI rendering issues across the desktop and web clients.
The /verify and /code-review skills are no longer run automatically by Claude. Users must now explicitly invoke them using /verify or /code-review commands when needed.
This release fixes multiple permission-check bypasses affecting Windows PowerShell, Bash, and remote sessions; adds the EndConversation tool for handling abusive users; improves diagnostics with progress heartbeats and structured logging; and resolves numerous stability issues including memory leaks, file encoding bugs, and background session management problems.
Version 2.1.212 introduces significant improvements to session management with /fork now creating background sessions, adds safeguards against runaway loops with search and subagent caps, and includes numerous bug fixes for stability, performance, and edge cases across workspace operations, streaming, and integrations.
Claude Code 2.1.211 adds support for forwarding subagent text and thinking in stream-json output, fixes critical security issues with permission preview character filtering and session management bugs, resolves model override and MCP server reconnection problems, and improves terminal performance and background agent reporting.
This release adds a live elapsed-time counter for tool calls, fixes critical issues with git isolation for worktree subagents, session crashes, and background worker stability, and improves permission handling, error messages, and agent UI interactions across multiple surfaces.
Fixed an issue where /model and other dialogs were incorrectly blocked in Claude Agents background sessions due to an overly broad guard. This reverts the problematic restriction.
This release adds accessibility features like screen reader mode and vim insert-mode remaps, introduces corporate launcher support via CLAUDE_CODE_PROCESS_WRAPPER, and fixes numerous stability issues including context window resets, memory leaks, slowdowns in permission-heavy sessions, and various tool failures.
This release makes auto mode available without opt-in on Bedrock, Vertex AI, and Foundry; fixes terminal freezing during streaming, remote settings consent dialogs, spurious prompt-injection warnings, and numerous issues with auto-updater, transcripts, worktrees, Remote Control sessions, and plugin configuration handling.
Claude Code 2.1.206 adds directory path suggestions for the /cd command, introduces a /doctor check for optimizing checked-in CLAUDE.md files, and improves git operations by auto-allowing pushes to configured remotes. This release includes numerous bug fixes for MCP server configuration, OAuth authentication, model picker display, keyboard input handling, and agent view management, plus improvements to code review quality and the agents view interface.
This release includes numerous bug fixes for session management, file handling, agent status display, and MCP server integration, along with improvements to auto mode, the agent view, and memory usage during updates. A new /doctor command provides comprehensive setup diagnostics.
Corrected an issue where hook events failed to stream during SessionStart hooks in headless sessions, preventing remote workers from being incorrectly idle-reaped while processing hooks.
This release adds login expiration warnings, improves background session stability and performance, fixes multiple crashes and stalling issues on macOS and Windows, and enhances the agent interface with better visibility and error messages.
97 more changes from Claude Code. Sign up to read the whole changelog.
Sign up freeGitHub or email — no card needed.