claude.ai connectors default to enabled with no per-server opt-in, fail loudly on missing auth

New sessions load all connected claude.ai MCP connectors (Slack, Atlassian, Salesforce, etc.) by default, regardless of whether the user actually has access to those target systems. This surfaces as auth/tool errors mid-session instead of failing safely.

The only mitigation is manually enumerating every unwanted connector in deniedMcpServers (settings.json), which doesn't scale as new connectors get added and requires knowing the exact display-name string for each connector in advance.

Requested: default-deny posture, with opt-in per connector. Alternatively, a working allowedMcpServers path for connectors: today serverName entries in allowedMcpServers reject names containing spaces or dots (e.g. "claude.ai Microsoft Learn"), and the docs recommend serverUrl instead, which isn't exposed to users for connectors, so there is no practical way to build an allowlist for connectors at all today.

Security default should be: everything off, explicit enable when wanted.