Claude docs changes for April 25th, 2026 [diff]

Executive Summary

  • New Rate Limits API allows admins to programmatically read rate limits for their organization and workspaces, enabling gateways and alerting systems to stay in sync without hardcoded values
  • PostToolUse and PostToolUseFailure hooks now include a duration_ms field exposing tool execution time in milliseconds
  • Several Claude Code settings (autoScrollEnabled, editorMode, showTurnDuration, teammateMode, terminalProgressBarEnabled) moved from ~/.claude.json to ~/.claude/settings.json, with three new settings added (prUrlTemplate, autoScrollEnabled documented in settings)
  • Microsoft 365 MCP server added as a built-in remote MCP connector for SharePoint, OneDrive, Outlook, and Teams access
  • Legacy Amazon Bedrock integration now has its own dedicated documentation page, separated from the modern Messages API Bedrock page

New Claude Code versions

2.1.116

Note: This version was previously documented; the following entry was added:

  • Claude Code and installer now use https://downloads.claude.ai/claude-code-releases as the download URL (changed from the previous Google Storage bucket URL)

Claude Code changes

Changed documents

agent-sdk/typescript [Source]

  • Added optional duration_ms field to PostToolUseHookInput and PostToolUseFailureHookInput types, exposing tool execution time in milliseconds [line 1116] [Source]
  • Added optional aliases field to the SlashCommand type [line 2260] [Source]

agent-teams [Source]

  • Updated reference for teammateMode setting: configuration is now stored in ~/.claude/settings.json rather than ~/.claude.json

cli-reference [Source]

  • --from-pr flag now accepts GitHub Enterprise PR URLs, GitLab merge request URLs, and Bitbucket pull request URLs in addition to GitHub PR numbers and URLs [line 57] [Source]

env-vars [Source]

  • New CLAUDE_CODE_HIDE_CWD environment variable: set to 1 to hide the working directory from the startup logo, useful during screenshares [line 96] [Source]
  • Updated ENABLE_TOOL_SEARCH documentation to note it is now also disabled by default on Vertex AI (not just non-first-party base URLs)

google-vertex-ai [Source]

  • Added explicit note that MCP tool search is disabled by default on Vertex AI because the endpoint does not accept the required beta header; all MCP tool definitions load upfront instead. Set ENABLE_TOOL_SEARCH=true to opt in

hooks [Source]

  • PostToolUse hook payload now includes duration_ms: tool execution time in milliseconds, excluding time spent in permission prompts and PreToolUse hooks [line 1200] [Source]
  • PostToolUseFailure hook payload also gains duration_ms [line 1256] [Source]

mcp [Source]

  • Added Microsoft 365 as a built-in remote MCP server, giving access to SharePoint, OneDrive, Outlook, and Teams [line 451] [Source]
  • Updated tool search documentation: MCP tool search is now also disabled by default on Vertex AI, and ENABLE_TOOL_SEARCH=true explicitly opts in for both Vertex AI and non-first-party base URLs
  • Removed outdated Windows-specific note about requiring cmd /c wrapper for npx-based MCP servers

monitoring-usage [Source]

  • OTel tool completion events now include tool_use_id, allowing correlation between OpenTelemetry events and hook-captured data [line 481] [Source]
  • OTel tool completion events now also include tool_input_size_bytes: size of the JSON-serialized tool input [line 488] [Source]
  • tool_use_id also added to OTel tool permission decision events [line 586] [Source]

permission-modes [Source]

  • Auto mode now also drops blanket PowerShell(*) allow rules on entry, the same way it drops Bash(*) [line 176] [Source]

plugin-dependencies [Source]

  • Clarified auto-update behavior for constrained dependencies: auto-update now fetches at the highest git tag satisfying all installed plugins' ranges rather than skipping the update entirely when the marketplace latest doesn't satisfy constraints [line 96] [Source]

settings [Source]

  • autoScrollEnabled, editorMode, showTurnDuration, teammateMode, and terminalProgressBarEnabled moved from ~/.claude.json (non-settings config) to settings.json starting with v2.1.119 [line 222] [Source]
  • New prUrlTemplate setting: URL template for the PR badge in the footer and tool-result summaries, allowing links to point at an internal code-review tool instead of github.com [line 197] [Source]

statusline [Source]

  • New effort.level field in status line JSON: current reasoning effort level (low, medium, high, xhigh, max), reflecting live mid-session /effort changes [line 154] [Source]
  • New thinking.enabled field: whether extended thinking is enabled for the session [line 155] [Source]

tools-reference [Source]

  • Removed the limitation note that auto mode does not work with the PowerShell tool (now supported)

API changes

New Documents

api/admin/rate_limits [Source]

API reference index for the Admin Rate Limits resource, providing an overview of the RateLimitListResponse model for organization-level rate limit queries.

api/admin/rate_limits/list [Source]

Full API reference for GET /v1/organizations/rate_limits. Lists all rate limit groups (model groups, batch, files, token count, skills, web search) with their configured limiter values. Supports filtering by group_type or by model string.

api/admin/workspaces/rate_limits [Source]

API reference index for the Workspace Rate Limits resource, describing the workspace_rate_limit response model which includes org_limit for reference comparison.

api/admin/workspaces/rate_limits/list [Source]

Full API reference for GET /v1/organizations/workspaces/{workspace_id}/rate_limits. Returns only rate limit groups that have workspace-level overrides; groups without overrides inherit organization limits and are omitted.

build-with-claude/claude-on-amazon-bedrock-legacy [Source]

New dedicated page for the legacy Amazon Bedrock integration using the InvokeModel and Converse APIs with ARN-versioned model identifiers. Covers authentication, bearer tokens, available model IDs with regional availability, global vs. regional endpoints, and PDF support. Claude Opus 4.7 is noted as available only through the modern Messages API endpoint.

build-with-claude/rate-limits-api [Source]

New guide for the Rate Limits API, providing programmatic access to rate limits for organizations and workspaces. Documents use cases (gateway sync, internal alerting, workspace audit), organization and workspace endpoints, filtering by group type, pagination, and an FAQ. Requires Admin API key.

Changed documents

agents-and-tools/remote-mcp-servers [Source]

  • Added Microsoft 365 as a built-in remote MCP server for SharePoint, OneDrive, Outlook, and Teams access

agents-and-tools/tool-use/build-a-tool-using-agent [Source]

  • Added cURL, CLI, and TypeScript code examples for the single-turn and multi-turn tool-use examples (previously Python only)

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

  • Updated recommended model to Claude Opus 4.7 (previously 4.6) and added multi-language code examples

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

  • Updated model references to Claude Opus 4.7 and added multi-language code examples

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

  • Updated model references to Claude Opus 4.7 and added multi-language code examples

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

  • Quick start example now uses Python code and multi-language tabs instead of a shell curl command; model updated to Claude Opus 4.7

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

  • Updated model references throughout from Claude Opus 4.6 to Claude Opus 4.7

api/admin [Source]

  • Added Workspace Rate Limits section documenting the new GET /v1/organizations/workspaces/{workspace_id}/rate_limits endpoint
  • Added Organization Rate Limits section documenting the new GET /v1/organizations/rate_limits endpoint
  • "managed" role removed from user and invite role enums
  • Usage report results now include account_id (user account that made the request) and service_account_id (OIDC-federation service account)
  • Cost report: added "session_usage" as a new cost_type value; removed the speed field (fast mode research preview)

api/admin/cost_report/retrieve [Source]

  • Added "session_usage" as a valid cost_type value in cost reports
  • Removed speed field (was a fast-mode research preview field) from cost report results

api/admin/invites/create [Source]

  • "managed" role removed from the list of valid invite roles

api/admin/usage_report/retrieve_messages [Source]

  • New account_ids filter parameter to restrict usage to specific user accounts [line 18] [Source]
  • New service_account_ids filter parameter to restrict usage to specific OIDC-federation service accounts [line 105] [Source]
  • New account_id and service_account_id options added to the group_by parameter [lines 72-74] [Source]

api/admin/workspaces [Source]

  • Added Workspace Rate Limits section with the new endpoint reference

api/rate-limits [Source]

  • Added reference to the new Rate Limits API for programmatically reading current rate limit configuration
  • Workspace rate limits section updated: removed the "Support for input and output token limits coming in future" note (now available); added note that workspace limits are set per limiter type; added Workspace limits are set per limiter type clarification

build-with-claude/administration-api [Source]

  • Added new "Rate limits" section linking to the Rate Limits API guide

build-with-claude/claude-in-amazon-bedrock [Source]

  • Updated link to legacy Bedrock integration to point to the new dedicated claude-on-amazon-bedrock-legacy.md page

managed-agents/memory [Source]

  • Fixed memory version listing code example: the paginated list result is now stored in a variable before iterating, and a version_id extraction example is shown

release-notes/overview [Source]

  • Added April 24, 2026 entry announcing the Rate Limits API

build-with-claude/workspaces [Source]

  • Added reference to the Rate Limits API for programmatic rate limit reads alongside the existing link to the rate limits documentation