Description
Claude.ai cloud MCP connectors (claude.ai *) periodically lose authentication in bulk — all connectors drop to "Needs authentication" at the same moment, not individually over time. There is no CLI mechanism to re-authenticate or restore them.
Evidence of bulk invalidation
The local mcp-needs-auth-cache.json shows 15 connectors all flagged within ~500ms of each other:
{
"claude.ai Canva": {"timestamp": 1776090864691},
"claude.ai Cloudflare Logpush": {"timestamp": 1776090864720},
"claude.ai monday.com": {"timestamp": 1776090864748},
"claude.ai Gmail": {"timestamp": 1776090864766},
"claude.ai Microsoft 365": {"timestamp": 1776090864769},
"claude.ai Notion": {"timestamp": 1776090864900}
}
This pattern points to a single upstream event (session rotation, token refresh failure) rather than individual provider token expiry.
Incomplete local tracking
claudeAiMcpEverConnected in .claude.json only tracks 5 of 21 cloud connectors:
["claude.ai Slack", "claude.ai Cloudflare Docs", "claude.ai Atlassian", "claude.ai Mermaid Chart", "claude.ai Sentry"]
It's unclear whether this incomplete cache contributes to connectors failing to recover.
No recovery path from CLI
Cloud connectors are not stored in local config (mcpServers in .claude.json / settings.json). When they lose auth or disappear entirely, there is no CLI command to re-authenticate, re-sync, or restore them. The only option is to re-add them through the claude.ai web UI one by one.
Expected behavior
- Cloud connector auth should be resilient — individual provider token expiry should be handled by silent refresh, and a single upstream event should not invalidate all connectors simultaneously.
claudeAiMcpEverConnected should be a complete list, or not exist at all — an incomplete cache is misleading.
- There should be a CLI recovery path (e.g.,
claude mcp sync or claude mcp reauth <name>) so users don't have to leave the terminal.
Environment
- Claude Code version: latest
- Platform: macOS (Darwin 25.4.0)
- 21 cloud connectors configured via claude.ai account
Related issues
Description
Claude.ai cloud MCP connectors (
claude.ai *) periodically lose authentication in bulk — all connectors drop to "Needs authentication" at the same moment, not individually over time. There is no CLI mechanism to re-authenticate or restore them.Evidence of bulk invalidation
The local
mcp-needs-auth-cache.jsonshows 15 connectors all flagged within ~500ms of each other:{ "claude.ai Canva": {"timestamp": 1776090864691}, "claude.ai Cloudflare Logpush": {"timestamp": 1776090864720}, "claude.ai monday.com": {"timestamp": 1776090864748}, "claude.ai Gmail": {"timestamp": 1776090864766}, "claude.ai Microsoft 365": {"timestamp": 1776090864769}, "claude.ai Notion": {"timestamp": 1776090864900} }This pattern points to a single upstream event (session rotation, token refresh failure) rather than individual provider token expiry.
Incomplete local tracking
claudeAiMcpEverConnectedin.claude.jsononly tracks 5 of 21 cloud connectors:It's unclear whether this incomplete cache contributes to connectors failing to recover.
No recovery path from CLI
Cloud connectors are not stored in local config (
mcpServersin.claude.json/settings.json). When they lose auth or disappear entirely, there is no CLI command to re-authenticate, re-sync, or restore them. The only option is to re-add them through the claude.ai web UI one by one.Expected behavior
claudeAiMcpEverConnectedshould be a complete list, or not exist at all — an incomplete cache is misleading.claude mcp syncorclaude mcp reauth <name>) so users don't have to leave the terminal.Environment
Related issues