Five new Clerk MCP server tools let agents list organizations and memberships and update existing users' names, usernames, external IDs, locale, timezone, Device Trust bypass, and public, private, or unsafe metadata. Existing connections may lack the organizations:read permission; running whoami shows whether it is missing, and reconnecting the MCP server grants it.
Clerk Changelog
npx @buildinternet/releases get clerk-changelogCustom session claims can now use session.created_at and session.expire_at shortcodes to read a session's age and expiry from its token, and a new Backend API PATCH endpoint sets a single session's expire_at between 5 minutes and 10 years out. Changes are recorded as session.expiry.updated events in Admin and Application Logs.
Application Logs in the Dashboard now show each sign-in or sign-up as one row with its event count and final completed or failed status, expandable to the individual events in order. GET /v1/logs returns grouped logs by default; pass view=flat to get one item per event.
GET /v1/logs now returns email events, filterable by type and recipient email, and the Dashboard's Application Logs page shows them alongside other event types. Email Logs filters now match exactly instead of by substring, blocked sends appear as Bounced or Failed, and Message ID shows the Clerk email ID when available.
An MCP server can give an agent a one-time link that opens a browser already signed in as the user.
Native sessions can now have a custom maximum lifetime between 5 minutes and 10 years, set separately from web browser sessions. The option is free in development but requires the Pro or Business plan in production, and changes apply only to sessions created after saving.
Directory Sync now sends six directoryGroup webhook events for group and membership changes, and two Backend API endpoints read a group and its members with cursor pagination. Directories can be paused from the Dashboard or via PATCH /v1/directories/{directory_id}, and group pushes now produce fewer user.updated events since only joined or left members are recorded.
A minimum version can now be set per iOS and Android app in the Clerk Dashboard; older builds have most requests rejected and get an update screen linking to the App Store or Google Play, with users staying signed in through the update. Enforcement requires Clerk iOS SDK 1.7.0, Android SDK 1.4.0, or @clerk/expo 4.15.0 or later, and apps on iOS SDK 1.0.0 through 1.6.x show an error instead of the update screen.
The GET /v1/logs endpoint is now public, letting you look up Application Logs events from your own code or an AI agent, filtering on the same event types as the Dashboard. The endpoint is rate limited, returns 429 too_many_requests over the limit, and can't be used for syncing or exporting logs.
The Business plan adds two Clerk Protect protections that cap sign-ups from one email domain per hour and bursts of sign-ups appearing to use the same shared infrastructure. Both limits apply across IP addresses and complement the existing per-IP sign-up limit.
Clerk for Agent Auth, built on open OAuth standards and the Personal Agent Protocol, is entering private beta. Agents that speak the protocol can discover an application, have their identities verified and authenticated by Clerk, and report feedback when they cannot complete a task.
Users who sign in with a password, social connection, or reset-password code can use a passkey instead of an SMS code or authenticator app as their second factor, including during reverification; supportedSecondFactors can now include passkey for custom flows. Requires @clerk/clerk-js 6.40.0 and @clerk/ui 1.41.0 or later, and the feature must be enabled in the dashboard for apps created before July 8, 2026.
With Test mode enabled, Clerk no longer sends any email to +clerk_test addresses, extending the existing skip beyond verification codes and new device notices to sign-in links, invitations, and receipts. Skipped emails are still recorded and trigger the email.created webhook with delivered_by_clerk set to false.
Clerk Protect adds configurable protections to sign-up and sign-in flows, including blocking VPN, Tor, proxy, and hosting network sign-ups, challenging residential proxies, country blocklists or allowlists, per-IP sign-up limits, and blocking risky email domains on the Business plan. Every application gets CAPTCHA challenges for suspicious sign-ups, Device Intelligence checks, account lockout after failed sign-ins, Device Trust, user enumeration protection, and blocking of email subaddresses and disposable addresses.
A new Backend API endpoint merges a duplicate user into another, moving email addresses, phone numbers, connected accounts, passkeys, TOTP, backup codes, organization memberships and metadata onto the kept user before deleting the duplicate. Merges are irreversible, so dry_run: true previews the result; a new user.merged webhook and Admin/Application Logs events record each merge, and merges are rejected for Directory Sync-managed users or when the deleted user has an enterprise SSO account.
Adding the submit_feedback tool from @clerk/mcp-tools to an MCP server lets agents report missing tools, unexpected behavior, and blockers as they hit them, and those reports can be read back through Clerk's MCP server. Setup is documented in the agent feedback guide.
Expo native components now ship in @clerk/expo-native-components, so apps not using them no longer bundle the Clerk iOS and Android SDKs or require iOS 17. The old @clerk/expo/native import path is deprecated and will be removed in the next major version; on Android, apps without the package no longer migrate biometric credentials enrolled with earlier SDK versions.
The API keys page in the Clerk Dashboard no longer displays the full value of existing Secret keys, showing only part of each key so it can be matched to an environment. Existing Secret keys keep working and do not need rotation; code snippets that need a Secret key can now add a new one without leaving the page.
The Clerk MCP server now connects through your MCP client's sign-in flow so an agent can create applications, configure authentication, manage users, and set up production instances alongside SDK snippets and quickstarts, scoped to what you approve and what your workspace role allows.
Enterprise SAML connections now trust up to five signing certificates at once, letting you add an IdP's next certificate before a key rotation so sign-ins keep working across the switch. Certificates are managed as a full list via idp_certificates on the Backend API or the Certificates list in the Dashboard, and the single idp_certificate field still works but is deprecated.


