Claude Code v2.1.186: MCP CLI Auth, Bash Auto-Response, and a Critical Subagent Permission Fix

TL;DR: Claude Code v2.1.186 adds CLI MCP auth, bash auto-response, and stops silently denying background subagent permission prompts.

Three meaningful changes in the June 22 v2.1.186 release:

claude mcp login / claude mcp logout. Authenticate MCP servers directly from the CLI without opening the /mcp menu. Pass --no-browser for headless or SSH environments — useful when the MCP server needs an OAuth flow but you're on a remote box.

Bash auto-response. !-prefixed bash commands now automatically trigger a Claude response after the command runs — no manual prompt needed to continue the workflow. Opt out per-project with "respondToBashCommands": false in settings if you prefer the old behavior.

Background subagent permission surfacing. Previously, when a subagent running in the background hit a permission prompt it couldn't auto-approve, it would silently deny and continue (or fail). Now those prompts surface in the main session so you can approve or deny them explicitly. This is a significant UX fix for anyone running unattended or scheduled subagent tasks — jobs were likely failing silently before.

Also fixed: streaming "Content block not found" errors after a machine went to sleep and woke up mid-session. And /review <pr> now uses the same engine as /code-review medium, so review quality is consistent across both entry points.

Why it matters: The subagent permission fix alone is worth updating for — if you run background agents that shell out or call external tools, they were probably auto-denying permissions and you didn't know.

Sources: Claude Code changelog, Releasebot claude-code, ClaudeCodeLog on X