Claude docs changes for June 23rd, 2026 [diff]

Executive Summary

• New claude mcp login/logout CLI commands let you authenticate MCP servers over SSH or without opening the interactive panel
• Shell mode ! commands now trigger Claude to respond automatically — eliminating the need for a second prompt after running commands
• Background subagents now surface permission prompts in your main session instead of silently auto-denying tool calls
• New plugin-relevance.md guide explains how marketplace operators can configure Claude Code to suggest plugins based on session signals
• Computer use is no longer HIPAA eligible (corrected in the data retention table)

New Claude Code versions

2.1.186

New features

• Added claude mcp login <name> and claude mcp logout <name> to authenticate MCP servers from the CLI without opening the interactive /mcp menu, with --no-browser stdin redirect support for SSH
• Added status filtering (press f) to the /workflows agent detail view
• Added a "Skills" section to the /plugin Installed tab
• Added teammateMode: "iterm2" setting for explicit iTerm2 native split panes via the it2 CLI
• Added "Claude Platform on AWS · refresh credentials" option to /login when awsAuthRefresh is configured
• ! bash commands now trigger Claude to respond to the output automatically; set "respondToBashCommands": false to keep the previous context-only behavior

Existing feature improvements

• Improved claude mcp get and claude mcp remove to suggest the closest configured server name on a typo
• Improved memory: the agent is now reminded to compact its MEMORY.md index when nearing the size limit
• Improved skill frontmatter: display-name, default-enabled, fallback, and metadata.* keys now accept kebab-case, snake_case, and camelCase
• Improved malformed SKILL.md YAML frontmatter handling: loads the skill body with empty metadata instead of failing silently
• Changed CLAUDE_CODE_MAX_RETRIES to cap at 15; for unattended sessions use CLAUDE_CODE_RETRY_WATCHDOG instead
• Changed background subagents to surface permission prompts in the main session instead of auto-denying; the dialog shows which agent is asking, and Esc denies just that tool
• Changed /review <pr> to use the same review engine as /code-review medium

Major bug fixes

• Fixed streaming requests failing with "Content block not found" or JSON parse errors after the machine wakes from sleep
• Fixed subagent transcript scroll position bleeding into the main transcript on exit
• Fixed Agent(type) deny rules and Agent(x,y) allowed-types restrictions not being enforced for named subagent spawns
• Fixed Workflow agent({schema}) subagents looping forever on repeated schema validation failures instead of aborting after 5 attempts
• Fixed ~~strikethrough~~ showing literal tildes instead of rendering as strikethrough
• Fixed session cost not showing for usage-based Enterprise and Team subscribers
• Fixed agent teams: teammates spawned via tmux/pane backends now inherit the leader's --effort level

Claude Code changes

New Documents

plugin-relevance [Source]

New guide for marketplace operators and enterprise administrators. Explains how to add a relevance block to a plugin's marketplace.json entry so Claude Code suggests the plugin based on session signals (working directory, CLI commands run, files read, host URLs, and manifest dependencies). Covers the three suggestion surfaces (spinner tip, session-start notification, Discover tab pin), how to enable suggestions via pluginSuggestionMarketplaces in managed settings, field reference for all signal types, and how to validate your marketplace with claude plugin validate.

Changed documents

agent-sdk/python [Source]

• CLAUDE_CODE_MAX_RETRIES is now capped at 15; new CLAUDE_CODE_RETRY_WATCHDOG=1 variable mentioned for unattended runs that need to retry capacity errors indefinitely. [line 826] [Source]

agent-sdk/typescript [Source]

• CLAUDE_CODE_MAX_RETRIES is now capped at 15; new CLAUDE_CODE_RETRY_WATCHDOG=1 variable mentioned for unattended runs. [line 464] [Source]

agent-teams [Source]

• Added iterm2 as a new teammateMode value (v2.1.186): uses iTerm2 native split panes via the it2 CLI and shows an error with the install command if it2 is missing. [line 94] [Source]

claude-platform-on-aws [Source]

• Removed the marketing banner/contact-sales block at the top of the page.
• Added documentation that with awsAuthRefresh configured, /login shows a "Claude Platform on AWS · refresh credentials" option that re-reads AWS credentials without restarting Claude Code. [line 40] [Source]

cli-reference [Source]

• Added claude mcp login <name> and claude mcp logout <name> commands to the reference table, including --no-browser flag for SSH use. [lines 25-26] [Source]
• Added iterm2 as a valid value for --teammate-mode (added in v2.1.186). [line 105] [Source]

commands [Source]

• Clarified that /review runs the same engine as /code-review on a GitHub pull request (not a deeper cloud-based review). [line 13] [Source]
• Updated /review [PR] description: now accepts a PR number, lists open PRs with no arguments, and uses the /code-review engine. [line 96] [Source]

env-vars [Source]

• CLAUDE_CODE_CONNECT_TIMEOUT_MS is now marked as removed in v2.1.186 (now a no-op); API_TIMEOUT_MS should be used instead. [line 148] [Source]
• Added CLAUDE_CODE_FORCE_STRIKETHROUGH env var (v2.1.186): forces strikethrough rendering when terminal supports it but is not auto-detected. [line 204] [Source]
• CLAUDE_CODE_MAX_RETRIES is now documented as capped at 15 (v2.1.186). [line 207] [Source]
• Added CLAUDE_CODE_RETRY_WATCHDOG: set to 1 in unattended sessions to retry 429/529 capacity errors indefinitely instead of failing after CLAUDE_CODE_MAX_RETRIES. Requires v2.1.186. [line 239] [Source]
• Updated CLAUDE_CODE_FORCE_SESSION_PERSISTENCE to note that the tmux case is auto-detected since v2.1.178 and to mention screen sessions as another use case. [line 203] [Source]

errors [Source]

• Added Waiting for API response · will retry in to the error lookup table. [line 35] [Source]
• Documented the 20-second stall detection threshold (v2.1.185+; was 10 seconds in earlier versions) that shows the "Waiting for API response" banner before any retry. [line 55] [Source]
• Added CLAUDE_CODE_RETRY_WATCHDOG to the retry tuning table. [line 62] [Source]

interactive-mode [Source]

• Shell mode ! description updated: Claude now responds automatically to command output (v2.1.186). [line 71] [Source]
• Added note explaining the new auto-response behavior, how to disable it with respondToBashCommands: false, and that before v2.1.186 shell mode always added output to context without a response. [line 270] [Source]

mcp [Source]

• Added new "Authenticate from the command line" section documenting claude mcp login <name> and claude mcp logout <name>, including --no-browser for SSH connections. [lines 473-491] [Source]

monitoring-usage [Source]

• Updated CLAUDE_CODE_MAX_RETRIES description to note the cap at 15. [line 967] [Source]

permissions [Source]

• Clarified that --add-dir loads both .claude/settings.json and .claude/settings.local.json (not just plugin settings). [line 320] [Source]

plugin-marketplaces [Source]

• Added relevance as a new marketplace-specific plugin field (requires v2.1.152+), with a link to the new plugin-relevance.md guide. [line 214] [Source]
• Clarified that some git servers (e.g., AWS CodeCommit) do not support fetching commits by SHA, so the ref must still exist on those servers. [line 248] [Source]

settings [Source]

• Added respondToBashCommands setting (v2.1.186): controls whether Claude responds after a ! shell command runs; default true. [line 261] [Source]
• Added iterm2 as a valid teammateMode value (v2.1.186). [line 276] [Source]

skills [Source]

• Added note that malformed YAML frontmatter causes Claude Code to load the skill body with empty metadata (so the skill still works via /skill-name) and that --debug reveals the parse error. [line 767] [Source]

sub-agents [Source]

• Updated background subagents behavior (v2.1.186): permission prompts now surface in the main session naming the requesting subagent; Esc denies that one tool call without stopping the subagent. Before v2.1.186, background subagents auto-denied any tool call that would have prompted. [line 674] [Source]
• Clarified that SendMessage is always available for resuming subagents by agent ID or name; structured team-protocol messages require agent teams to be enabled. [line 762] [Source]

tools-reference [Source]

• Updated SendMessage description: no longer requires CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 for resuming subagents by agent ID; structured team-protocol messages still require agent teams. [line 32] [Source]
• Updated background subagent tool permission behavior description to match v2.1.186 changes. [line 91] [Source]

ultrareview [Source]

• Updated comparison table to include /code-review as a distinct option: now shows all three commands (/code-review, /review <pr>, /code-review ultra) with their targets, depth, duration, cost, and best-use. [lines 82-92] [Source]

workflows [Source]

• Added f key shortcut to filter the agent list in the selected phase by status in the /workflows progress view. [line 96] [Source]

API changes

Changed documents

agents-and-tools/tool-use/define-tools [Source]

• Updated model recommendation from Opus 4.7 to Opus 4.8 for complex tools and ambiguous queries. [line 9] [Source]
• Added a full multi-language code example (with language tabs) for forcing tool use with tool_choice, replacing the previous single-line snippet. [line 138] [Source]

agents-and-tools/tool-use/handle-tool-calls [Source]

• Added search_result as a valid content block type for tool results (alongside text, image, and document). [line 30] [Source]

agents-and-tools/tool-use/strict-tool-use [Source]

• Added a "Next steps" card section linking to web fetch tool, tool use with prompt caching, structured outputs, and define tools. [line 121] [Source]

api/cli/beta/messages/batches/results [Source]

• Page restored from a "Console temporarily unavailable" error state to the full CLI reference for retrieving Message Batch results, including parameters and response schema.

build-with-claude/batch-processing [Source]

• Added "Next steps" cards linking to search results and prompt caching.

build-with-claude/extended-thinking [Source]

• Restructured supported models list into a table showing budget_tokens support, recommended thinking mode, per model. [line 101] [Source]
• Added Claude Fable 5 and Claude Mythos 5 to the interleaved thinking support table; added Claude Haiku 4.5 (not supported) and clarified earlier Claude 4 models. [line 217] [Source]
• Updated max output tokens: Claude Sonnet 4.6 now supports 128k (was 64k). [line 144] [Source]
• Replaced per-model comparison text with a condensed table covering budget_tokens, thinking output, interleaved thinking, and block preservation. [line 242] [Source]
• Added "Adaptive thinking" to the "Next steps" section. [line 267] [Source]

build-with-claude/fast-mode [Source]

• Added that fast mode is compatible with streaming (where the OTPS gain is most visible). [line 39] [Source]
• Clarified that fast mode does not silently fall back to standard speed on rate limits — it returns a 429 or 529 instead. [line 316] [Source]

build-with-claude/handling-stop-reasons [Source]

• Added a quick-reference table at the top mapping each stop_reason value to when it occurs and what to do. [line 12] [Source]
• Added stop_details: null to the example response for end_turn. [line 44] [Source]
• Clarified that model_context_window_exceeded is currently only typed in the SDKs' beta namespace and requires client.beta.messages; added which models support it natively vs. via a beta header. [line 236] [Source]
• Added "Next steps" cards linking to refusals/fallback, tool runner, streaming, and errors. [line 512] [Source]

build-with-claude/refusals-and-fallback [Source]

• Changed the primary code example from TypeScript to multi-language tabs (Python shown by default). [line 20] [Source]

build-with-claude/search-results [Source]

• Added "Next steps" cards linking to citations, web search tool, Messages API reference, and prompt caching. [line 537] [Source]

manage-claude/api-and-data-retention [Source]

• Computer use is no longer HIPAA eligible — changed from "Yes" to "No" in the HIPAA column of the feature eligibility table. [line 175] [Source]

test-and-evaluate/strengthen-guardrails/handle-streaming-refusals [Source]

• Replaced guidance to reset context with recommendation to retry on a fallback model using server-side fallback or SDK middleware. [line 47] [Source]
• Added "Migration notes" section covering: refusals arrive as HTTP 200 (not errors), newer models include stop_details, retry on a different model, check batch results for refusals, and centralize on stop_reason. [line 105] [Source]
• Added "Next steps" cards. [line 128] [Source]