TL;DR
Fast mode silently switched to Opus 4.7. If your fast-mode workflows depend on Opus 4.6 behavior, set
CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE=1to pin to the old model.MCP_TOOL_TIMEOUTwas not actually raising the fetch timeout for remote HTTP/SSE servers. Tool calls were capped at 60 seconds regardless of the configured value; if you set a higher timeout and still hit 60s walls, this is your fix (more below).claude agentscan now configure sessions at dispatch time. New flags--add-dir,--settings,--mcp-config,--plugin-dir,--permission-mode,--model,--effort, and--dangerously-skip-permissionslet you customize background agents without separate configuration.Background agents no longer disappear after macOS sleep or binary upgrades. The daemon now detects clock jumps and exits cleanly when the binary path changes, fixing two classes of crash-loops (more below).
New in 2.1.142
2.1.142 (May 15, 2026)
- Added new
claude agentsflags:--add-dir,--settings,--mcp-config,--plugin-dir,--permission-mode,--model,--effort, and--dangerously-skip-permissionsto configure dispatched background sessions - Fast mode now uses Opus 4.7 by default (previously Opus 4.6). Set
CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE=1to pin fast mode to Opus 4.6 - Plugins with a root-level
SKILL.mdand noskills/subdirectory are now surfaced as a skill - The
/plugindetails pane andclaude plugin detailsnow show LSP servers a plugin provides /web-setupwarns before replacing an existing GitHub App connection- Fixed
MCP_TOOL_TIMEOUTnot raising the per-request fetch timeout for remote HTTP and SSE MCP servers, which capped tool calls at 60 seconds regardless of the configured value - Fixed background sessions not recognizing pre-existing git worktrees, blocking Edit while EnterWorktree refused to create a duplicate
- Fixed background sessions disappearing and daemon reconnect failing after macOS sleep/wake; the daemon now detects clock jumps instead of treating them as elapsed idle time
- Fixed daemon not exiting cleanly after the binary is upgraded (e.g.
brew upgrade), causing dispatched agents to crash-loop on the deleted path - Fixed background agents crash-looping when the Claude-in-Chrome extension is connected without a shared tab
- Fixed clicking links in an attached
claude agentssession; the background worker's headless browser shim no longer applies while attached - Fixed
claude agents"v to open in editor" using the daemon's default editor instead of your shell's$EDITOR/$VISUAL - Fixed
claude agentsdeadlocking on Windows with network-drive working directories; Ctrl+C now works during startup - Fixed background-color bleed when attaching to a
claude agentssession from Apple Terminal or other 256-color-only terminals - Fixed
claude --bg --dangerously-skip-permissionsnot persisting across retire/wake - Fixed session titles being derived from the URL when the first message is a link
- Fixed redundant
set_modelrequests from remote clients injecting duplicate/modelbreadcrumbs into the transcript - Fixed plugins using
skills: ["./"]showing a false "path escapes plugin directory" error - Fixed plugin cache cleanup deleting the active plugin version directory when no installation metadata is present
- Fixed
/pluginbrowse pane showing "0 installs" for newly published plugins - Fixed plugin advisories not naming every
plugin.jsonkey that shadows a default folder - Improved reactive compaction: the first summarize attempt now seeds from the original request's overflow size, avoiding a wasted near-full-context retry
- Improved hook configuration error: configuring a prompt- or agent-type hook for
SessionStart/Setup/SubagentStartnow shows a clear "use a command-type hook instead" error - Removed stale
/model claude-sonnet-4-20250514suggestion from Usage Policy refusal messages
Notes
MCP_TOOL_TIMEOUT was a no-op for remote servers
MCP_TOOL_TIMEOUT has been the documented way to raise per-tool timeouts for slow MCP servers since 2.1.105. For stdio (local) servers it worked. For remote HTTP and SSE servers, the configured value was never forwarded to the underlying fetch, so every call hit a hard 60-second ceiling. If you gave up on the setting and worked around it with smaller payloads or request splitting, you can revert those workarounds now.
Background agent stability after the agent view launch
2.1.139 introduced claude agents as a research preview three days ago. This release patches three daemon-level issues that made long-running agents unreliable. The sleep/wake fix (clock-jump detection) addresses disappearances reported on macOS laptops. The upgrade fix stops the old daemon from hanging around after brew upgrade or npm update, which left newly dispatched agents pointing at a deleted binary. The worktree fix resolves a deadlock where background sessions could not edit files if a worktree already existed at the target path.
Together these make claude agents substantially more viable for workflows that span hours or survive a laptop lid close.