
Plugin Auto-Connect Launches for Claude Users
Users can now auto-connect plugins when logging in to Claude, using MCP servers that accept Claude as a trusted issuer, removing the need to re-authorize on each session.

Users can now auto-connect plugins when logging in to Claude, using MCP servers that accept Claude as a trusted issuer, removing the need to re-authorize on each session.
Team and Enterprise admins can now configure plugins once, and teammates are connected automatically on first use, eliminating repeated per-user sign-in flows.
Enterprise-managed auth (EMA) lets admins connect plugins organization-wide through their identity provider, built on an open MCP extension, so individual members never have to search for or authorize plugins themselves.
When an organization's identity provider doesn't support enterprise-managed auth, Claude itself can act as a trusted issuer for signed-in members, letting them auto-connect to plugins after a one-time admin setup.
October brought mods, multi-session improvements in the desktop app, better usage visibility, and a range of smaller quality-of-life updates. The full recap is in the linked changelog thread.
Long idle sessions are now automatically compacted just before the prompt cache expires (around 1 hour), so returning to them later only requires reprocessing a summary rather than the full conversation. Pro and Max users also get a warning before resuming a very long conversation that would eat a large chunk of their 5-hour limit.
Typing 'ultrathink' in a message temporarily bumps thinking effort to xhigh for just that turn without changing the session-wide setting. Subagent effort can also be set independently from the parent session. The /usage command now surfaces plain-language explanations of what's consuming limits, with a one-key fix when available.
Shell commands that hit their timeout no longer get killed; they keep running in the background instead. Also in this thread: the new CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS=1 flag for containers that want zero permission interruptions.
A new env var, CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS=1, lets throwaway containers bypass even the last few risky rm prompts that persist in normal bypass mode. Also new: cloud sessions let you click any diff line to leave a comment that gets sent with your next message, MCP servers can now ask follow-up questions mid-task and run long jobs in the background, and Max/Ultra/Team plans get a monthly API credit for building with the Claude API and Agent SDK.
The October changelog is live. The thread that precedes this post covers sessions sidebar improvements, inline diff comments, monthly API credits for paid plans, and MCP background job support.
An Anthropic team member updates their view: rather than everyone building permanent personalized tools with AI, the more likely future is ephemeral, task-specific software backed by a single persistent data store. The advice for builders: prioritize interoperability.
An Anthropic team member describes managing schema drift in AI-generated throwaway tools through built-in guardrails, sensible defaults, and deploying additional agents to handle maintenance.
Developers can add auto-connect support to their plugins today via published documentation, with the beta available on Team and Enterprise plans across 14 partner integrations including Figma, Notion, and Supabase.