Curated tools for Claude Code, without the noise.

Auto mode becomes the default in Claude Code for Pro, Max, and Team plans · Starting on August 14, new sessions on Pro, Max, and Team plans will run in auto mode. Instead of permission prompts, auto mode routes each tool call through a classifier targeted at blocking actions that are irreversible, destructive, or aimed outside your environment. Users can switch modes with Shift+Tab in the CLI or the mode dropdown on the desktop app, and admins can pin an org-wide default with defaultMode or turn auto mode off entirely with disableAutoMode.

Run Claude Code sessions on your own compute with self-hosted environments · Now in public beta, self-hosted environments let you run Claude Code sessions on your own infrastructure. You operate runners in either fixed or on-demand mode, and each session runs in its own checkout so work stays isolated between developers and accounts. Repository checkouts, build artifacts, and secrets stay on infrastructure you provision, while the conversation is sent to Anthropic for inference. Available to organizations on Claude Team and Enterprise plans.

Bringing MCP 2026-07-28 to Claude — a stateless core, standardized extensions, and hardened auth · MCP 2026-07-28, the fifth spec release of the Model Context Protocol, is live. It moves MCP from a bidirectional stateful protocol to a stateless request/response core so servers can deploy on serverless and edge infrastructure, graduates MCP Apps and Tasks into a versioned extensions framework, and aligns authorization with production OAuth 2.0 and OIDC deployments. Support is rolling out across Claude products.

Claude Opus 5 · Anthropic introduced Claude Opus 5 on July 24, 2026, for complex agentic coding and enterprise work. It is priced at $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8, with the model ID claude-opus-5, a 1M token context window, and 128k max output tokens. Adaptive thinking is now on by default, effort runs up to max, and two betas ship alongside it: mid-conversation tool changes and automatic fallbacks. It is the new default model on Claude Max and the strongest model on Claude Pro.

Ask Claude about the Anthropic Economic Index · A new connector lets you query the Anthropic Economic Index in claude.ai in plain language. Ask questions like "Which occupations use AI the most?" to explore AI adoption, task automation, and geographic usage patterns. There's nothing to install, but you do connect it once in the claude.ai connectors settings.

Claude Code 2.1.234 · Claude Code now continues your session automatically when a claude.ai usage limit resets, and you can turn it off in /config. Repos with a GitLab remote and an authenticated glab CLI get an MR !N badge in the footer and statusline, a new selection:clear keybinding action clears an in-app text selection, and CLAUDE_CODE_PROJECT_DIR_NAME lets hosts pick a short name for the per-project transcript directory. Loading the built-in claude-api skill drops from ~200k+ tokens to ~25k, and /permissions and /add-dir can now be opened while Claude is working. Over 30 fixes land across permission previews and credential masking, Remote Control sync, and queued ! shell commands.

Claude Code 2.1.233 · The two Bash permission changes from yesterday's 2.1.232 — permission checks on input redirections (`< file`) and on Cygwin-style symlinks on Windows — are reverted, and the auto mode regression that made ordinary `cd <dir> && <command> > file` commands stop for manual approval on Windows is fixed. GitLab support now extends to merge request URLs in the `--worktree` flag and the `claude agents` view, where MRs display as `!N`. Three opt-in settings land: `CLAUDE_CODE_TOOL_MEMORY_LIMIT` for memory cgroup limits on Bash tool commands on Linux, a `forward_user_identity` apps gateway setting that sends the signed-in user's identity as headers, and `CLAUDE_CODE_WEBFETCH_CACHE_TTL_MS` for the WebFetch session URL cache TTL. Todo/task-tracking tools are no longer available on Opus 4.8, Sonnet 5, Fable 5, Mythos 5, and newer models.

Claude Code 2.1.232 · Subagent forking is now enabled by default: a `subagent_type: "fork"` subagent inherits the full conversation and prompt cache, and non-teammate agent spawns in interactive sessions run in the background by default. Type `@` in the prompt to mention another Claude session by name, and interactive sessions on one machine now keep unique names. Plugin marketplaces accept bare `gitlab.com` repo URLs the way they accept `github.com` URLs, and GitLab token families are now redacted. The fixes cluster around permission bypasses and Remote Control: a PowerShell bypass through `$PSDefaultParameterValues`, a Windows bypass through Cygwin-style symlinks followed by Git Bash, and nested git repositories inheriting trust from a parent directory — each repository now requires its own trust confirmation.

Claude Code 2.1.229 · Plugin marketplaces can now come from a `command` source: a local command prints the plugin directory, which is re-resolved each session and applied without a restart. Gateway streaming responses send SSE keepalive pings during long thinking pauses, preventing idle-timeout disconnects on Vertex and Bedrock upstreams, and self-hosted runner sessions now get server-supplied hooks, matching managed-environment behavior. Several crashes are fixed — a RangeError in very narrow terminal windows, a crash on non-string `glob`, `file_path`, or `command` values, and a Windows crash on extended-length or UNC paths — along with long responses partly disappearing while streaming and being printed twice. `/commit-push-pr` no longer auto-approves git/gh commands carrying dangerous flags like `--force`, `--amend`, or `--no-verify`.

Claude Code 2.1.228 · Skills synced from claude.ai are hardened: they no longer shadow local commands or MCP prompts, their descriptions are sanitized and labeled, and on your machine their bodies don't run ! commands or expand @ files. Two fixes stop things from being deleted that shouldn't be — session cleanup deleting contents inside a project's memory folder, and background plugin-cache cleanup deleting a plugin's cache when its only version is a symlinked development checkout. Also fixed: interactive sessions that could stop redrawing entirely while the process kept running, git and Git Bash not being found on Windows, and self-hosted runner sessions ending early.