Claude restores web and API service, reports skipped scheduled runs

Anthropic closed a Claude outage affecting its apps, API and Claude Code at 15:38 UTC on November 13. Shortly beforehand it said some scheduled runs had been skipped and others could start more than two hours late.

A separate Microsoft 365 connector incident remained open. Shortly after closing the wider Claude outage, Anthropic reported signs of recovery for the connector and said some SharePoint searches still returned errors.

A sign-in incident came first

Anthropic had already responded to a sign-in incident earlier on November 13. It began investigating authentication failures on the Claude website and in Claude Code at 09:58 UTC. The incident record also listed the developer console and Claude Cowork as affected. The company deployed a fix at 11:05 UTC and marked the incident resolved at 11:58 UTC.

Anthropic opened the second incident at 13:09 UTC. Its investigation expanded to Docs, Sheets and Managed Agents. At 14:36 UTC the web app and API were showing signs of recovery while connectors were still having trouble.

By 15:33 UTC, error rates on the main services had normalized and existing connections to Docs and Sheets worked again. Adding new connectors on the website was still temporarily unavailable. About two and a half hours elapsed between the first investigation notice and the resolution.

Scheduled runs had delays and gaps

Scheduled tasks in Cowork run saved instructions on a chosen timetable. Anthropic’s help center gives examples such as a daily briefing drawn from emails and calendar events or a weekly summary based on a connected spreadsheet. The incident record did not say how many runs were skipped or whether they were automatically replayed afterwards.

In Claude Cowork, each execution of a recurring task creates a separate session. Anthropic’s help center says the Scheduled page shows upcoming and past runs, with the output available inside each completed session. Cowork normally runs those tasks remotely. The help center says tasks set up before October 6 on Pro, Max or Ultra that already run on the user’s computer continue to run there. A team with a recurring briefing can have the schedule on its list while the record of that day’s run and the briefing itself are tracked in the execution history.

Claude Code’s cloud scheduler, previously called routines, also provides a task history. It distinguishes runs that finished, failed or were cancelled. Each row opens the session that performed the work. The documentation says a session can end successfully even when a connector was unavailable or a request failed during the run. Those details appear in the session transcript.

Some SharePoint searches still returned errors

The Microsoft 365 connector incident began on November 12 at 17:52 UTC, with errors in requests such as mail and calendars and trouble setting up new connections. Anthropic attributed the connector disruption to a Microsoft service incident that day. It also reported problems with the Claude add-in for Outlook.

For work and school accounts, the connector gives Claude access to connected Outlook mail and calendars, OneDrive files, SharePoint and Teams. For users who link a personal Microsoft account, the connector covers Outlook and OneDrive. For a work account, it can search only material that the person already has permission to see in Microsoft 365. A question can reach Claude while a separate request to retrieve the company document it needs still fails. SharePoint search covers material held in sites and document libraries, so those errors can interrupt tasks that depend on finding a specific file.

At 15:40 UTC on November 13, Anthropic reported signs of recovery for the Outlook add-in as well as the connector. At the 19:44 UTC check, it listed the API, Claude Code and the developer console as operational. It showed degraded performance for the Claude website and Cowork, and the connector incident with the remaining SharePoint search errors was still open.