v2.1.186

Source: Anthropic Claude Code Date: 2026-06-22 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.186

Summary

Claude Code v2.1.186 (2026-06-22) is a large maintenance and capability release. Key additions: claude mcp login/logout <name> for CLI-based MCP server authentication (including --no-browser for SSH flows); ! bash commands now auto-trigger Claude responses (opt-out via respondToBashCommands: false); background subagents now surface permission prompts in the main session rather than auto-denying. Numerous UX fixes across the agents panel, /workflows, skill loading, and session cost display.

Implications

Feeds floor-fencing autonomy on two dimensions:

• Credentials/MCP auth: mcp login/logout moves server authentication out of the interactive menu and into a scriptable CLI path, including SSH support. This is a provenance and credential-management improvement — agents operating over SSH can now authenticate MCP servers without a browser, closing a gap in unattended operation.
• Permission legibility: surfacing background subagent permission prompts in the main session (with agent identity shown) directly improves the reversibility/legibility fence — operators can see and deny tool requests from named agents rather than having them auto-denied silently.
• The Agent(type) deny-rule fix (enforcement for named subagent spawns) strengthens agent governance: scope restrictions that were previously bypassable on named spawns now apply correctly.