Small releases, big everyday relief. Claude Code 2.1.186 ships a detail that anyone working a lot with MCP servers will appreciate — especially those doing it remotely over SSH.
MCP auth without the menu detour
Until now, signing in to an MCP server meant going through the interactive /mcp menu. With this release you can do it straight from the command line:
claude mcp login <name>— signs you in to an MCP serverclaude mcp logout <name>— signs you back out
Sounds unspectacular, but in practice it’s a real workflow difference. You stay in your shell’s flow, script the authentication if you want, and never have to pop open the interactive UI.
The actual gem is the --no-browser flag: it lets you complete the login flow via stdin redirect even when there’s no browser around — say, on a remote machine you’ve SSH’d into. That’s exactly where MCP auth used to be a pain. On top of that there’s status filtering (press f) in the /workflows agent detail view, so you can quickly narrow long lists down to what you actually care about.
2.1.187: cleaning up after killed agents
Right behind it came 2.1.187 with a fix most people won’t notice — but anyone who’s been stuck in a broken Git worktree will value. When an agent was aborted or killed, locked registrations were left behind under .git/worktrees/. Those orphaned entries are now cleaned up automatically. Fewer mysterious worktree corpses, less manual tidying.
My take
These aren’t headline features — but they’re exactly the kind of update that makes the tool feel grown-up. The more MCP servers you connect, the more often you hit auth friction; the more parallel agents you run, the more likely one leaves a mess in a worktree. Tackling both shows where Anthropic is polishing Claude Code: precisely where power users bump into it every day. Browserless SSH login is one of those things you only miss once you’ve had it — and then never want to give up.
Sources: Claude Code Release v2.1.186, Claude Code CHANGELOG