You can now turn off MCP task support by setting CLAUDE_CODE_DISABLE_MCP_TASKS. This release has 440 entries, and 284 of them are usable today. Accounts offered Life Sciences use cases get a new /use-case command to pick one. That choice is saved for later sessions. A new app:closeSession shortcut action lets you bind a key to close a session shown under session tabs. Programs that launch Claude Code can pass working directories exactly as given with the new permissions.workingDirectoriesAsGiven setting.
There are 12 entries in the build that are not switched on yet. A Wrap up tool would let Claude end a turn with a short recap and up to two next-step buttons. Claude Code now tracks messages for taking back a sent message with Esc, but the take-back does nothing yet. Hooks of type agent and mcp_tool can be read, but a remote allow list keeps them off and is empty by default. A spot is reserved for a live preview sharing dialog, with no way to open it yet.
Updates on macOS no longer delete the app bundle a running session is using. Saving settings no longer overwrites a config file that appeared while Claude Code was creating its own. Cancelling the theme picker no longer quits Claude Code. SDK hooks set to block on failure now stop the action as intended. Reading an unchanged file again no longer returns a short "file unchanged" note. The CLAUDE_CODE_LAZY_ENV_GETTERS environment variable is no longer recognised.
Written by our agent from the shipped bundle, not by Anthropic.
Adds CLAUDE_CODE_DISABLE_SUBAGENT_HANDBACK and CLAUDE_CODE_DEEP_TWILIGHT, and lists PARSED_WILLOW and DAPPER_CUPCAKE in more allowlists
ExtensionsA new ui.backdrop function lets plugins draw a blurred, tinted backdrop, and connecting clients can say they draw a Backdrop
ExtensionsClaude Code now has an app:closeSession shortcut action and a SessionTabs keybinding context for closing a session shown under session tabs
InterfaceRemote sessions gain new CLAUDE_CODE_* settings, such as a stall-check interval and a slow-skills watch, and CLAUDE_CODE_LAZY_ENV_GETTERS is dropped
/use-case command to choose a Life Sciences use caseAccounts offered Life Sciences use cases get a /use-case command and a startup dialog to pick one, saved for later sessions
Want the reasoning? Read walks the 51 entries that probably matter to you, each one opening to what changed and why.
Read this release → Every row →Anything you can use today, anything that visibly changes, and anything worth poking at. One line each, open for detail.
Adds CLAUDE_CODE_DISABLE_SUBAGENT_HANDBACK and CLAUDE_CODE_DEEP_TWILIGHT, and lists PARSED_WILLOW and DAPPER_CUPCAKE in more allowlists
Unclear What Claude Code does differently with a prompt it treats as a mid-turn host prompt is not stated.
Hooks gain agent and mcp_tool types, accepted only when the remote tengu_woven_ladder list names them
A new ui.backdrop function lets plugins draw a blurred, tinted backdrop, and connecting clients can say they draw a Backdrop
Unclear Whether plugin function hooks are switched on by default, and whether the terminal shows the backdrop, is not clear.
Claude can attach local files to Gmail drafts and messages with your approval, and HIPAA or ZDR policies can block connector file sending
Unclear Whether Gmail attachments are switched on when the server sends nothing is not known.
Claude Code now has an app:closeSession shortcut action and a SessionTabs keybinding context for closing a session shown under session tabs
Unclear It is not clear whether session tabs can be reached yet or what pressing the close shortcut does.
A new Wrap up tool lets Claude end a turn with a short recap and up to two buttons offering what to do next
Unclear Which users meet the two other conditions needed for the Wrap up tool to appear is not known.
A new wrap-up feature is controlled by CLAUDE_CODE_ENABLE_WRAP_UP and defaults on for remote sessions with prompt suggestions on
Unclear What the wrap-up feature actually does is not stated.
Remote sessions gain new CLAUDE_CODE_* settings, such as a stall-check interval and a slow-skills watch, and CLAUDE_CODE_LAZY_ENV_GETTERS is dropped
Unclear It is not clear what the check does once it is turned on.
A new Esc take-back feature for rewinding sent messages is wired in, but its rewind steps are still empty, so nothing happens yet
/use-case command to choose a Life Sciences use case#Accounts offered Life Sciences use cases get a /use-case command and a startup dialog to pick one, saved for later sessions
CLAUDE_CODE_FIZZY_GROVE setting overrides how an interrupt is handled#Setting CLAUDE_CODE_FIZZY_GROVE now overrides a remote switch that changes what happens when work is interrupted
Unclear It is not known what the interrupt behaviour actually does differently when it is turned on.
A new ReadWidgetContext tool lets Claude read the latest state that an MCP App view has shared with Claude Code
Unclear Whether the ReadWidgetContext tool is available by default, or only when something switches it on, is not shown.
File sends through connectors get a connectorFileSend permission that bypass mode does not skip, with a dialog listing the files and refusing unreadable lists
Unclear Which connectors bring up this dialog, and whether it is switched on for everyone, is not stated.
SDK hook callbacks can set onFailure to block so a failed check stops the action, and the SDK refuses to start if Claude Code cannot honour it
claude edit-permission-rules now accepts compose requests, refuses certified rules, and catches duplicate rules that differ only in spelling
Plugins that draw their own screens through ui.render gain a new place to draw, called Backdrop
Unclear Whether anything in Claude Code shows the Backdrop spot yet is not stated.
A version of the tool for leaving plan mode makes the plan optional, to be left out when the plan is kept in a Claude Doc
Unclear It is not clear whether this version of the tool is the one in use, since the captured stock description did not change.
In plan mode, Claude can write its plan as a Claude Doc beside the chat, and approving it means Claude carries out the Doc as written
Unclear What switches Doc-based plans on, and whether they are available to anyone yet, is not stated.
Claude Code can recognize a docs connector's tools and build claude.ai/code/artifact links in remote sessions, behind a remote switch
CLAUDE_CODE_DISABLE_MCP_TASKS#Setting CLAUDE_CODE_DISABLE_MCP_TASKS turns off MCP task support, and the remote switch for it can now default to on in some setups
Unclear Which setups get the on-by-default fallback is not clear.
CLAUDE_CODE_PLAIN_DELTA_DIRECT_WRITE setting for faster stream-json output#Setting CLAUDE_CODE_PLAIN_DELTA_DIRECT_WRITE with the fast-path option writes streamed text pieces straight out when the reader is waiting
permissions.workingDirectoriesAsGiven setting for programs that launch Claude Code#A host program can now pass working directories spelled exactly as given, via --settings, without Claude Code following links
New /design-sync messages say a design system has moved, or is moving, to Claude, that updates can't be sent there yet, and how to open it
If a session belongs to another account, Claude Code sends nothing and asks you to send a message to choose which account to use
Unclear What causes this message to appear, and whether it is switched on for everyone, is not clear.
The cloud home screen now appears inside a new frame with tabs and an optional roster when that frame is available, and keeps its old layout otherwise
Claude Code now has a place reserved for a live preview sharing dialog, with nothing yet showing how it opens
Unclear It is not clear whether the live preview share dialog can be opened by anyone yet or what it shows.
This build adds internal pieces for remote sessions and for notices when promotional credit is running low or used up
Unclear It is not clear which of these pieces do anything for readers yet.
New code can build narrow always-allow rules for a command, host or MCP tool, refusing risky cases and saving to local or project settings
Unclear Whether anything in Claude Code offers these rules to you yet is not stated.
mcp_authenticate gains redirectUriOnly, which fails sign-in instead of falling back to a localhost listener when the custom redirect is rejected
The operation that updates permission rules now accepts compose alongside add and remove
Unclear What the compose operation does to a permission rule is not stated.
SDK hosts can send ultrathink_typed to say a person typed "ultrathink", so the turn runs at xhigh effort when the session allows it
Unclear What makes system/init report ultrathink_typed_raises as true is not shown, so it is unclear when a host can use this.
Depending on how Claude Code was launched, it clears CLAUDE_CODE_ENTRYPOINT and may set CLAUDE_CODE_HOST_SHORT_TEMP_DIRS to 0
Unclear It is not clear which launch types are affected or what CLAUDE_CODE_HOST_SHORT_TEMP_DIRS changes.
To deliver a message, Claude Code can now move running tool calls to the background without stopping them, keeping their original time limit
Unclear It is not clear whether this behaviour is switched on for anyone in this release.
When a search service is available, the WebSearch tool sends searches through it and reports how many results domain filters left out
Unclear It is not clear what makes the search service available, so it is unknown which setups get this routing.
Without an API key, Claude Code now refuses to connect to the API with a session_owner error while a session-owner hold is in place
Unclear What puts a session-owner hold in place is not shown, so it is unclear when this error would appear.
The Claude API reference that ships inside Claude Code now includes a page on message threads
Release checks can now report a failure for the macOS app bundle of Claude Code, not only for updates
Claude Code now saves a list of session owners with the transcript and can let a session's first turn skip the check against overlapping turns
Unclear Whether session owners are used by any feature you can reach yet is not stated.
Behind a switch that defaults to off, Claude Code can record the email and organization of accounts you use and label the current one
Behind a server-side switch, Claude Code running headless in the cloud keeps up to 256 tool states, or 1 MiB, sent along with a message
The opening message in stream-json output gains client_thinking and flag_restate_noop fields for Cowork cloud workers, plus two new refusal reasons
Claude Code can now check whether a connected desktop app says it draws a backdrop
Unclear Whether anything a user can see uses this check yet is not stated.
The part of Claude Code that runs hooks can now run hook types other than shell commands, where before it ran only command hooks
Unclear It is not clear when Claude Code runs hooks through this path for non-command hooks, so whether your own hooks are affected is unknown.
Claude Code now has a widget context hint, a new kind of background note it can attach to a conversation
Resuming a session now reads a recorded session owner and, when it asks about the owner, skips the stale-resume prompt and does not reconnect Remote Control
Unclear It is not settled exactly when Claude Code decides to ask about the session owner, so it is unclear which resumes this affects.
Self-hosted runners now skip trusting the workspace when a new server-config check passes, and the --trust-workspace help describes a stricter gate
Unclear It is not clear what condition makes a self-hosted runner skip saving trust.
A plugin's .mcp.json or mcpServers entry can now refer to a resource of kind mcp-servers instead of a file
Unclear How a plugin provides an mcp-servers resource is not described.
displayName#An MCP server's settings can now include a displayName, and a badly formed value is ignored instead of breaking the config
For some tools, an allow rule ending in (*) or () now means any input, behind a remote switch that is on unless turned off
Unclear Which tools this applies to, and how the old reading differs in practice, is not stated.
What would fix it?
Smaller changes and internals, grouped as the pipeline found them. Nothing is dropped, it is only further down.
14 more of these are in What probably matters to you, above.
The background task panel and footer can now show workflow runs happening in the cloud alongside local ones
Unclear It is not stated which sessions get the new counting.
Artifact listing now pages through results and includes a shared list alongside your own, falling back to the old list when paging is unavailable
When a command pipes into tail, Claude Code can save the full output to a file and tell Claude to read it instead of re-running
Unclear What the second output-handling path controlled by the same switch does is not known.
When a command pipes into tail, Bash keeps a saved copy of the full output and tells Claude where it is; background shells can also add an end note
/goal no longer sets, clears or shows the goal when the message was relayed from Slack, Teams or a project thread; send it yourself instead
In headless and SDK use, mcp_set_servers and mcp_reconnect requests now go through a queue beside the main loop instead of running inline
The routine tool now asks before creating, updating or running a routine, and flags connectors the server attached that Claude did not ask for
Claude Code can report its sandbox state to a host app, including support, locks, restrictions and which managed-settings sources apply
MCP image, binary and large results now report whether each was shown, saved, kept, cut or failed to save, so hosts can see how it was handled
When a model needs auto or plan mode, the plan approval choices drop the bypass-permissions option and the default-mode option, and offer auto mode
Unclear Which models need auto or plan mode, and so get the shorter list of choices, is not stated.
Reading a resource an MCP server lists as a skill resource can now be allowed without asking you first
Unclear Which permission setting decides whether the prompt is skipped is not clear.
The cloud workflows dialog can now open directly on a given workflow run, and stopping a run now shows a result message
Unclear It is not clear what opens the dialog on a specific run.
The sandbox now guards more folders inside .git against replacement and can make the git remotes and branches folders read-only
Account plugins' HTTP hooks can now be run on another machine with headers and allowedEnvVars, and long plugin listings are cut to fit
Admins can block /remote-control through managed settings, its status message no longer carries an upgrade prompt, and a follow-on tip is switchable
A built-in check now asks you first when a web or shell tool call would include your signed-in account's email address
Unclear Whether the server-side switch this check also depends on is on for any given account is not known.
MCP tools that can send this session's files elsewhere now need your permission before Claude uses them
Unclear It is not stated which MCP tools count as able to send files elsewhere, or whether this applies to every MCP tool.
Resuming a conversation started under another login now opens a dialog to exit, change login, or copy it to your current account
Unclear It is not clear whether this check is switched on for everyone or depends on a setting.
File references in MCP tool calls now work beyond Drive, and calls are refused if they changed after approval or were not made directly
Unclear It is not clear which MCP tools use the recipient checks or whether any of them are available yet.
A new notice says your own hooks aren't running because a host kept them out, setup did not finish, or of launch mode, and Claude relays it
HTTP hooks defined in project settings may now go through a separate sender, and some failure messages say "the hook's server" instead of the URL
Unclear When project hooks take the separate path is not clear.
The usage credit message now names the covered models from your account's limit details, falling back to Fable, and its date range is fixed
A label listing which plans can use a feature now reads 'Individual, Team, or Enterprise' instead of 'Pro, Max, Team, or Enterprise'
Claude Code has two new messages for when a computer cannot run local tools for a cloud session because one side's version is too old
A note tells Claude that the session's GitHub proxy creates private repositories only, and to keep -F private=true
Claude Code reworked its checks for the xhigh effort level, and resets the session's effort when the picked model cannot use that level
Unclear Whether the reworked checks change when xhigh effort is actually offered to you is not settled.
When an action on another device fails because one side's Claude is too old, Claude now says which side to update
Claude's instructions now allow games with similar mechanics and spell out that it will not copy a game's name, characters, art, music, levels or look
DesignSync now refuses paths claude.ai/design cannot store cleanly and uses the stored names in plans and listings
Self-hosted runners pass agentProxyUrl to the server, park finished code-review sessions, and report their capabilities differently
The native installer now removes old app bundles, keeps locked versions, and can keep every bundle when the launcher is managed by something else
Claude Code now keeps system cross-certificates a server's certificate chain needs instead of dropping every looping one, and adds an opt-out
Unclear It is not settled exactly what setting CLAUDE_CODE_DISABLE_LOOPED_CROSS_CERT_DROP switches off.
A subagent's result can now list its task-list calls in a new taskCalls field, controlled by a switch renamed to tengu_bright_kettle
Self-hosted runners now hand CLAUDE_CODE_EXIT_AFTER_STOP_DELAY from the server to sessions and gain checks for restricted and host-managed runs
Unclear It is not clear what being marked as a host-managed runner child changes for a session.
The SDK warning sent when a requested model is refused now carries a model_restriction field naming both models and when it happened
In auto mode, a refused subagent action now shows its own warning, such as "N agent actions not run", and repeated warnings are grouped
In remote mode, cancelling now gives a take-back step a chance to restore your drafts, and rewinding keeps several drafts at once
Unclear What the take-back step does in each case when you press Esc in remote mode is not shown.
When you entered plan mode from auto mode, choosing plain "yes" on the exit-plan dialog now puts you back in auto mode
Unclear It is not clear what makes a model need auto or plan mode, which is the second case that leads to auto mode.
claude -p now reports session data lost to a full disk#If a full disk or used-up quota stops Claude Code saving session data in print mode, it now reports this when the run ends
Before creating a design Artifact, the question form now asks at most four questions and offers three looks drawn as colour cards
Unclear Who can use the Artifact form, and whether it is switched on for everyone, is not clear.
When a session is resumed in plan mode, its plan document is now restored as well as the mode
Unclear Whether this is switched on for everyone or depends on a switch is not clear.
CLAUDE_CODE_SHELL#When the shell's bash is too old to run a command reliably, Claude Code refuses it and suggests setting CLAUDE_CODE_SHELL to zsh or a newer bash
Unclear Which bash versions trigger the refusal, and whether it is switched on for everyone, is not stated.
Choosing a cloud workflow in the footer can now send a request to stop it, with a message if stopping fails
Some background hooks that print plain text and succeed now have that text passed to Claude as context when a session starts
Unclear It is not stated which hooks have this behaviour turned on, or whether you can turn it on for your own.
Grep returns partial matches instead of failing on a timeout, and Grep and Glob results now carry a timedOut flag
Behind a server switch, the sandbox no longer follows a write-protected symlink that points at another host or cannot be safely checked
Unclear What this switch does when the server sends nothing is not settled, so it is unclear whether the new handling applies to you by default.
Behind a new remote switch, Claude Code can check a shell command, refuse it before bash runs, and stop on old bash versions
Unclear It is not clear whether the switch counts as on or off when the server sends no value.
In some plugin sessions, logs and errors for hooks and the status line show a display name or "command left out" instead of the command
Unclear It is not stated which plugin sessions use the hidden display name.
In remote environments, Claude is now told to commit or push on the other machine only if the project folder there is a git checkout
The footer summary of background tasks was reworked so monitor tasks are counted, while a set of only workflows still reads as dynamic workflows
In one mode, the collapsed 'In progress…' line for a subagent now shows only the ctrl+o hint, without tool-use and token counts
Unclear Which mode hides the counts is not clear.
Published verbatim by Anthropic for v2.1.317. Text is unmodified from the upstream changelog. Everything else on this page came out of the bundle instead, which is why the two lists don't match.
Of these 87 bullets, 14 name something an entry on this page also names, 26 name something no entry here does, and 47 name nothing specific enough to line up either way. The pairings are made on names both sides wrote down, a flag or a setting or a slash command, so read one as probably the same thing rather than as a fact, and read the middle number as candidates rather than as a miss count.
/schedule when a new cloud scheduled task gets connectors you did not ask for, with how to remove them
No entry names this onFailure: "block" for Agent SDK hook callbacks, so a callback that fails blocks the action instead of letting it run
No entry names this anthropic.plan, anthropic.seat_type, anthropic.provider and anthropic.custom_base_url to OTel metrics, events and spans
No entry names this GET /v1/organizations/analytics/cost, for reading past spend over a date range in total, per developer, per day or per IdP group
No entry names this admin: set, kept for admin.identity_retention_days, which the cost report groups by
No entry names this CLAUDE_CODE_DISABLE_NESTED_SESSION_ID: set it to 1 on a self-hosted runner to have claude -p run inside a session use that session's ID and transcript, as in earlier versions
Probably new-session-writes-lost-system-message-and-full-disk-exit-ha, nested-session-id-opt-out-env-var-added claude auth status exiting before a slow token refresh was saved, which could sign you out on the next run
No entry names this #95822Short-lived commands (`claude auth status`, `claude --bg …`) start an OAuth refresh at init and exit without saving it, leaving the profile with a spent refresh token Open
effortLevel setting being ignored by sessions that an app starts ahead of time and then opens in the project, such as new sessions in the desktop app's Code tab
No entry names this #104336[BUG] Desktop app (Windows): Code-tab sessions sometimes start at Medium effort, ignoring the High pick and the project's effortLevel Open
#93735Project-scope effortLevel in .claude/settings.json is never applied in desktop-app (Code tab) sessions; effective effort varies per session with no settings change Open
claude --from-pr reporting "No conversations found" in projects with many sessions when the linked session was older than the first page
No entry names this /loop wakeups sometimes not being scheduled when Claude left a field out of its request
No entry names this /schedule leaving a cloud scheduled task's next run time and link out of its results after creating or updating it
No entry names this claude -p exiting 0 when a full disk or quota lost its transcript or ~/.claude.json writes; it now prints an error and, in text and json output, exits 1. Full NFS exports may be missed
Probably new-session-writes-lost-system-message-and-full-disk-exit-ha --setting-sources
Probably added-folder-plugins-follow-setting-sources, runner-passes-empty-setting-sources-when-the-arg-is-explic /team-onboarding guide when asked to delete one without naming which
No entry names this /team-onboarding guide was deleted when the server refused the delete
No entry names this AWS_BEARER_TOKEN_BEDROCK when their env config sets it to empty; the variable is now unset, so the server can use the machine's own AWS credentials
No entry names this DISABLE_PROMPT_CACHING_OPUS and DISABLE_PROMPT_CACHING_SONNET having no effect on [1m], other-region and undated spellings of the default model
Probably prompt-caching-disable-env-vars-now-compare-models-by-identi /artifacts and the Artifact tool's shared list showing no shared artifacts when you have 200 or more of your own
Probably get-artifacts-route-added-to-an-allow-list https:// endpoint through a proxy never being sent when they started while another export was still in progress
Probably only-https-origins-accepted-for-url-origin-helper, fetching-the-servers-certificate-chain-to-repair-tls-verifi http:// endpoint
No entry names this OTEL_EXPORTER_OTLP_TIMEOUT up to 30 seconds
Probably otlp-proxy-tunnel-agent-and-timeouts /remote-control answering "Unknown command" instead of saying that Remote Control is turned off by the disableRemoteControl setting
Probably remote-control-status-message-drops-upgradenudge, remote-control-slash-command-can-be-blocked-by-managed-polic claude -p run inside a cloud or self-hosted runner session sharing that session's ID: it wrote into the session's transcript, and a restart could resume the wrong conversation
Probably new-session-writes-lost-system-message-and-full-disk-exit-ha claude -p run inside a cloud or self-hosted runner session generating a title of its own and sending it to that session
Probably new-session-writes-lost-system-message-and-full-disk-exit-ha cloud_credential_error
No entry names this /design-sync to say up front when your design system has moved to Claude, and stop before building instead of failing at upload
Probably design-sync-messages-for-design-systems-that-moved-to-claud --chrome --debug) ahead of --resume
Probably resume-hint-can-include-launch-flags-file-writes-gain-creat plugin-authoring skill to offer once, after a mod you built works, to help share the mod
No entry names this /usage showing a developer's own spend limit as 100% used after their group's pooled budget ran out
No entry names this CLAUDE_CODE_RESUME_INTERRUPTED_TURN_MAX_AGE_MS
No entry names this claude remote-control a fork of "Name (fork)" is "Name (fork 2)"
Probably fork-name-fallback-for-session-titles-removed chat.fontSize is raised
No entry names this ultrathink not raising effort for its turn or showing its rainbow when telemetry is off
Nothing to match on claude remote-control, when you sent it within a minute of your previous message
No entry names this A model matched these bullets to the GitHub issues they fix, so a link can be wrong.
1 prompt changed.
Claude Code, interactive mode
6 prompt changes in this release could not be quoted from the build, so no entry on this page describes them.
263 documentation changes were recorded within 24 hours either side of this release, nearest first. The closest 12 are below. They're here because they happened near this release in time. That's not a claim that this release caused the edit, or that the page documents anything in it.
The 69 literal strings found in the bundle, with the number of entries that name each one. Picking one searches for it. A name is here because this build's code mentions it, which is not the same as it working or being finished.
What's wrong with this entry?