Claude Code 2.1.327: stdin input for MCP setup, a /status diagnostics list, and tighter permission checks
Anthropic's changelog for Claude Code 2.1.327 (2026-11-17) adds stdin input for claude mcp add-json and /status diagnostics, and changes permission behavior for symlinks, working directories and auto mode.
News date: About 6 min read
Contents
Claude Code is Anthropic's coding assistant for the terminal and editors. In its changelog, Anthropic lists version 2.1.327 (2026-11-17), with three main themes: a new way to configure things, a diagnostics list to help troubleshoot problems, and many tightened permission checks. A permission check is the mechanism by which Claude Code asks you or denies outright before running a command or reading or writing a file. The changelog does not mention plan tiers, regions or staged rollout details.
What new features does this update add?
An MCP server is a component that connects Claude Code to external tools. According to Anthropic's changelog, the new claude mcp add-json name - reads the server's JSON configuration from standard input, meaning content passed in through a pipe or paste. This way a token (an access credential, similar to a password) does not have to be written on the command line. However, this option is not available inside a Claude Code session. /status also gains a list showing broken agent, skill, command and output style files, keybinding errors, unusable sandboxes (isolated environments that limit what commands can access), ignored permission settings, and hook matchers that can never match.
- /install-github-app adds a confirmation step before changing a repository; /config shows an explanation when another setting overrides a row.
- A new attributionProject setting names the session's billing group for Claude apps gateways that have admin.billing_group_tag set.
- /tui classic is added as the name for the non-fullscreen renderer, and /tui default still works.
- VS Code: the Explorer right-click menu adds "Claude Code: Add to Chat", which adds selected files and folders to the chat as @ mentions.
What changed in permissions and security behavior?
The changelog's "Changed" section is mostly about tightened permissions. Path checks now reject some symlinks, which are shortcut files pointing to other locations. The "inside the working directory" determination has also changed. The working directory is the project folder Claude Code is allowed to operate in, so some paths now require permission. permissions.blockReadsOutsideWorkingDirectories now asks about more Bash commands in bypass and auto modes. Auto mode is the mode where a safety check decides on the user's behalf whether to allow an action. In auto mode, this setting also denies some commands. Separately, the setting now refuses more Bash commands that name a path under a Read or Edit deny rule, where it used to ask.
In addition, Bash commands whose variables expand to very long text are denied, and the PowerShell tool now asks first for one more command. Claude in Chrome refuses to upload some credential files, which the user must attach on the page themselves. Artifact calls are denied once Artifacts is turned off in settings, including in sessions that are already running.
| Item | Change described in the changelog | Likely impact on users |
|---|---|---|
| Symlinks and working directory | Some symlinks are rejected; the in-working-directory check changed | Some paths now prompt |
| ExitWorktree | Asks first when there may be unsaved work; denied when no one can answer, unless a rule or mode allows it | May be blocked in unattended situations |
| auto mode | Ignores more allow rules that run arbitrary packages, such as Bash(pnpm dlx *), leaving the decision to the safety check | Such allow rules no longer take effect directly |
| Billing-classifier fallback | A notice explains an upcoming release will remove the fallback | Affected actions will be asked about or denied |
| Very long Bash variables | Commands expanding to very long text are denied | Use shorter values or files instead |
| Claude in Chrome | Refuses to upload some credential files | Attach them yourself on the page |
How does this affect everyday workflows?
A subagent is a helper that the main session dispatches to handle a subtask. Subagents started without a specified model now keep their original model when resumed, including after running /model. On Amazon Bedrock, Google Vertex AI and Microsoft Foundry, subagent progress summaries in terminal sessions are now written from short notes, which the changelog says uses far fewer tokens. Non-interactive (-p) mode runs fewer git commands before the first request.
On the interface side, the arrow keys in most permission prompts now stop at No instead of cycling. For a short time after resuming with fg, most prompts ignore answers other than Esc. On the VS Code side, unused tools in the IDE connection (openFile, saveDocument, getOpenEditors and related items) were removed.
Frequently asked questions
Which plans or regions does 2.1.327 apply to?
Anthropic's changelog only lists the changes and does not describe plan tiers, regions or staged rollout details. Refer to Anthropic's official changelog and documentation.
What are the limits of the new claude mcp add-json usage?
According to the changelog, claude mcp add-json name - reads the server's JSON from standard input so that tokens need not appear on the command line. This option is not available inside a Claude Code session.
What will change for auto mode later?
The changelog says auto mode's notice now explains that an upcoming release will remove the fallback for billing-classifier requests. After that, each action that would have been checked will be asked about or denied.
How does the new /status list help?
Per the changelog, it lists broken config files, keybinding errors, unusable sandboxes, ignored permission settings and hook matchers that can never match, which helps with troubleshooting configuration problems.
Browse the latest news in this topic
Found this guide helpful? Share it with someone who needs it
Signal Blocks Screenshots by Default on Android and iOS 27 to Stop AI Agents Like Siri and Gemini From Reading Your Screen
The Verge reports that Signal is restricting screenshots on Android and iOS 27 to keep OS-level AI agents from reading message content. Users can turn screenshots back on in settings, but Signal warns this weakens its privacy protections.
Claude Code 2.1.329: Code Mode Works With Telemetry, /tasks Lists Scheduled Tasks, macOS Computer Use Gets Stricter Safety Rules
Kimi Code CLI v2.7.0 Adds a Native Amazon Bedrock Provider: Use Your AWS Credentials Instead of a Separate API Key
AWS Rounds Up 13 SageMaker Inference Features: Managed NVIDIA Dynamo on HyperPod, MIG GPU Partitioning, Cross-Region Routing
Sources
- 2.1.327Checked:
Compiled from the references and verification dates listed above; fares, schedules, product features, and terms can change, so please confirm the latest details on the relevant official website before travelling, booking, or purchasing.