Claude Code v2.1.206 Major Updates

This is Ishikawa from the Cloud Business Division. Claude Code v2.1.206 (released 2026-07-09) has been released. Rather than featuring prominent new functionality, this is primarily a maintenance-focused release centered on stabilizing authentication, MCP, background execution, and session resumption — addressing pain points that tend to cause friction in day-to-day operations.

Update Summary

v2.1.206 includes 27 changes: 9 new features and improvements, and 18 bug fixes, focusing on practical issues commonly encountered around authentication, MCP, and agent management. There are no breaking changes or deprecations. Personally, I'm particularly interested in the /doctor updates.

Notably, today (around 4 AM on 2026/07/10), the 5-hour and weekly rate limits for all users were reset. Users who had reached the weekly rate limit for Claude Fable 5 can use it again until 7/12.

https://x.com/ClaudeDevs/status/2075279141352706215

Highlighted Updates

• New Feature: /doctor now includes a check that suggests slimming down committed CLAUDE.md files by removing content that Claude can already infer from the codebase.
• Bug Fixes: Fixed an issue where all models were failing with incorrect errors instead of prompting /login when a login expired; fixed an issue where claude --resume / --continue was unresponsive to keyboard input at startup; fixed an issue where MCP server request_timeout_ms was being ignored and timing out at 60 seconds.
• Security (Safety / Accident Prevention): EnterWorktree now requires confirmation before entering a git worktree outside the project's .claude/worktrees/.

Update Details

New Features

• Added CLAUDE.md slim-down check to /doctor: Running /doctor now includes a check that suggests removing content from CLAUDE.md that Claude could infer just by reading the codebase. This does not automatically delete or rewrite anything — it only provides "suggestions" for reduction candidates, with the final decision left to human review.

  For example, the following types of "readable from code" descriptions may become reduction candidates:

  • Directory structures and lists of key files (what's in src/, tests/, etc.)
  • Dependency libraries and frameworks used (discoverable from package.json, etc.)
  • Build, test, and lint commands (simple listings like npm test, npm run lint)
  • Language/framework versions and formatting rules readable from linter/formatter configs

  On the other hand, content that cannot be inferred from code alone — such as "why things are done a certain way," like keeping an old response format for backward compatibility, or not running a particular process synchronously due to high production data volumes — represents project-specific intent, constraints, and gotchas that should be retained.

  Since CLAUDE.md is loaded into context every turn, the benefits of reducing redundant descriptions are meaningful in practice. It can reduce unnecessary token consumption, prevent truly important instructions from getting buried in lengthy text, and avoid the dual-maintenance and staleness issues that arise when CLAUDE.md descriptions diverge from actual files like package.json after code changes.

• Added directory path suggestions to /cd: /cd now displays directory path suggestions, matching the behavior of /add-dir.

• Gateway: /login now supports public gateway endpoints: /login now supports Anthropic-operated public gateway endpoints. This is an improvement for users accessing Claude Code via Gateway.

Improvements

• /commit-push-pr now auto-permits pushes to configured push remotes: In addition to origin, git push to the configured push remote for the repository (remote.pushDefault, or the sole remote if only one exists) is now automatically permitted.
• Improved update timing for background agents: Background agents now update to the new version in the background immediately after a Claude Code update, instead of performing a slow stale session upgrade at attach time.
• Improved finding quality for /code-review: The quality of /code-review findings has been improved in claude-opus-4-8 across all effort levels.
• Improved display and deletion behavior in the agents view: The status column now uses the full terminal width instead of being truncated at 64 characters. Additionally, sessions are no longer displayed twice, and deleted background jobs remain deleted. Note that Ctrl+X has been changed to permanently delete completed sessions, so users accustomed to the previous behavior should take note.

Safety / Accident Prevention

• Added confirmation prompt to EnterWorktree: Confirmation is now required before entering a git worktree outside the project's .claude/worktrees/ directory. This is a safety measure to prevent unintentionally entering an unexpected working tree.

Fixes (Key Items)

• Fixed misleading error display for expired logins: Fixed an issue where, when a login expired, all models would fail with the misleading error "There's an issue with the selected model" instead of prompting the user to run /login.
• Fixed keyboard input unresponsiveness with --resume / --continue: Fixed an issue where claude --resume and --continue were unresponsive to keyboard input at startup.
• Fixed per-server timeout being ignored for MCP: Fixed an issue where MCP servers configured via --mcp-config or .mcp.json were ignoring per-server request_timeout_ms settings, causing long-running MCP tool calls in new sessions to time out at the 60-second default.
• Fixed cold-start crash with --permission-prompt-tool: Fixed an issue where, when --permission-prompt-tool pointed to an MCP server, a cold start before the server had finished connecting would crash with "MCP tool not found."
• Fixed unnecessary re-authentication for OAuth MCP servers: Fixed an issue where OAuth-authenticated MCP servers would require manual re-authentication after just a single token refresh failure.
• Fixed "running" freeze in desktop sessions: Fixed an issue where desktop sessions would freeze on "running" after a slash command was sent mid-turn.
• Fixed startup hang on Bedrock: Fixed an issue where startup would hang for several minutes when using the awsCredentialExport helper on a network with restricted egress.
• Additionally, numerous minor bugs have been fixed, including /model picker pricing and line display, usability of the agents view and workflow detail display, /status and /doctor display, LSP plugin hint display, and key input on Windows.

Closing

v2.1.206 is a release that steadily stabilizes the key aspects of daily operation: authentication, MCP, background execution, and session resumption. Among the new features, the /doctor suggestion for slimming down CLAUDE.md looks practically useful, and the fixes address issues that tend to halt work when encountered — such as MCP timeouts and unresponsive keyboard input with --resume. If any of these sound familiar, why not update and give it a try?

References

https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md

https://code.claude.com/docs/en/changelog