Claude Code MCP auth controls

Claude Code v2.1.186 added command-line MCP login and logout commands and shipped fixes around background-agent permission prompts, allowed tool types, and repeated workflow-schema failures. Why it matters: MCP-enabled agents are only practical when authentication, permissions, and failure loops are inspectable. CLI-level auth controls and permission fixes reduce friction for safer agent operations.

Try this: Before connecting MCP tools to an agent, verify the login/logout path, allowed tool list, denied actions, and behavior when a background task requests permission.