Anthropic: Claude Code v2.1.136 brings 54 fixes, MCP OAuth fix and hard-deny rule
Anthropic released Claude Code v2.1.136, a maintenance release with 54 changes that introduces the new settings.autoMode.hard_deny rule for unconditionally blocking actions in auto mode, fixes the MCP OAuth race condition that forced users to re-login daily, and resolves an API 400 error during extended thinking.
This article was generated using artificial intelligence from primary sources.
Anthropic released Claude Code v2.1.136 on May 8, a maintenance release that according to official release notes contains 54 changes. The focus is on security rules in auto mode, MCP authentication stability, and fixes around extended thinking.
What does the new hard_deny rule change?
The most significant addition is settings.autoMode.hard_deny, a rule category in the auto mode classifier that unconditionally blocks actions, regardless of user intent or existing allow exceptions. It is a hard boundary above existing permissions, intended for actions an organization never wants to see executed automatically. In addition, the CLAUDE_CODE_ENABLE_FEEDBACK_SURVEY_FOR_OTEL variable was added, which brings back a session quality survey for enterprise installations using OpenTelemetry collection.
How were MCP authentication issues resolved?
The community’s most painful complaint — daily re-logins to MCP servers — was resolved by fixing a race condition during simultaneous OAuth token refresh. When multiple remote MCP servers requested a refresh in parallel, tokens were overwriting each other and being lost. A rare login loop where concurrent credential writes could overwrite a freshly rotated token was also fixed. A long-standing bug was resolved: MCP servers configured in .mcp.json, plugins, and claude.ai connectors no longer disappear after /clear in VS Code and JetBrains extensions and the Agent SDK.
Other important fixes
An API 400 error was resolved when extended thinking emits a redacted thinking block after a tool call — a typical case in longer agentic chains. WSL2 users can now paste images from the Windows clipboard via the new PowerShell fallback when xclip or wl-paste fail. --resume and --continue no longer crash for projects with underscores in the path, and plan mode now correctly blocks file writes even when a matching Edit(...) allow rule exists. A series of minor display issues was fixed (CJK characters, scrollback, ReasonML diff), along with bash output and markdown block issues.
Frequently Asked Questions
- What does the new settings.autoMode.hard_deny rule do?
- The rule unconditionally blocks certain actions in auto mode, regardless of user intent or existing allow exceptions. It is designed as a hard boundary that the classifier cannot bypass.
- Which MCP OAuth bug was fixed?
- A race condition during simultaneous token refresh from multiple MCP servers was losing the refresh token and forcing daily re-logins. The fix also resolves a rare login loop caused by concurrent credential writes overwriting a freshly rotated token.
- How does image paste in WSL2 now work?
- A PowerShell fallback was added that retrieves the image from the Windows clipboard when xclip or wl-paste cannot read binary data in the WSL2 environment.
Related news
AWS: Halliburton AI assistant for seismics cuts workflow creation time by over 95 percent
GitHub: Five risks and a 10-minute framework for reviewing AI pull requests
arXiv:2605.04012: SymptomAI in the Fitbit app with 13,917 patients outperforms independent clinicians in differential diagnosis