# Claude Code 2.1.321 fixes MCP token leak and two Bash sandbox and permission bugs

> Anthropic's changelog lists fixes for MCP OAuth secrets sent to another host on redirect and a read-only Bash command that ran outside the sandbox.

- Status: official (From the company, authors or authority themselves, with a link to their own page.)
- Topic: Safety & Security
- Published: 2026-11-09T00:00:00+00:00
- Updated: 2026-11-09T21:09:03.333001+00:00
- Web page: https://ai-news.ikangai.com/story/claude-code-2-1-321-fixes-mcp-token-leak-and-two-bash-sandbox-and-permission-bug/
- Seen in: Claude Code changelog
- Change: security (Security)
- Takes effect: 2026-11-09
- Applies to: Claude Code 2.1.321, MCP OAuth, Bash

## What changed

Claude Code 2.1.321 stops sending MCP OAuth secrets such as refresh tokens to another host when a server redirects them, and fixes a read-only Bash command that ran outside the sandbox without a permission prompt.

Anthropic's Claude Code 2.1.321 changelog, dated November 9, 2026, lists 74 bug fixes. Two of the first ones it names concern credentials and sandboxing. One fixes MCP OAuth requests that sent secrets such as refresh tokens to another host when a server redirected them. The other fixes a read-only Bash command that ran outside the sandbox without a permission prompt. The changelog gives no further detail on either, so the scope of each fix is known only from its one-line description.

The changelog says Claude Code no longer sends MCP OAuth secrets, such as refresh tokens, to another host when a server redirects them. The entry does not say how many users or servers were affected, or which versions before 2.1.321 were involved.

The changelog describes a read-only Bash command that ran outside the sandbox and raised no permission prompt. It does not say how the command was classified. A separate entry covers Bash reads such as grep -r foo "$PWD/src", which asked for permission when the same command with the path typed out ran without asking.

Two more entries concern the permission checks themselves. One fixes Bash permission checks that judged a file write from the wrong folder and so did not prompt. Another fixes deny rules and protected-file prompts failing to apply to the file that a cp or mv into a directory actually writes. The changelog also says that cp and mv spelled with a path, such as /bin/cp, now get the same path checks as plain cp and mv.

Other changes in the same release touch permissions as well. The changelog says mods that Claude writes will turn on without asking in auto and bypass modes, even with managed allowUnsandboxedCommands: false or allowManagedDomainsOnly. It also says mods' own tool calls raise no permission prompt, even under your own ask rules, though deny rules and organization policy still apply.

## Why it matters

Teams running Claude Code with MCP servers or sandboxed shell access should treat these as security-relevant fixes, since they cover credential exposure and sandbox bypass.

## What you can do

Upgrade Claude Code to 2.1.321 on machines and runners that use MCP OAuth or Bash permission rules.

## Original source

- [Claude Code changelog - Claude Code Docs](https://code.claude.com/docs/en/changelog)
- [claude-code/CHANGELOG.md at main · anthropics/claude-code · GitHub](https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md)

---

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it.

Cite as: IKANGAI AI News, "Claude Code 2.1.321 fixes MCP token leak and two Bash sandbox and permission bugs", 2026-11-09, https://ai-news.ikangai.com/story/claude-code-2-1-321-fixes-mcp-token-leak-and-two-bash-sandbox-and-permission-bug/
