[BUG] MCP Tool Permission System Broken in Claude.ai Web Interface (I cannot find anywhere to post this to get any eyes on it?)

Issue Summary

The MCP tool permission system in Claude.ai web interface has been completely non-functional for several days. When Claude attempts to execute MCP tools that require user confirmation, no permission dialogs appear and tools execute automatically without user approval.

Impact on MCP Ecosystem

This issue is critically blocking MCP adoption for business and enterprise use cases where user control and approval workflows are essential.
Multiple developers and users are affected:

• Reddit reports of similar permission system failures:

https://www.reddit.com/r/ClaudeAI/comments/1mlviq3/claude_desktop_filesystem_extension_stops_asking/

• Enterprise MCP servers cannot demonstrate proper user control during client demos
• Business applications requiring approval workflows are completely unusable
• Users lose trust when AI appears to have unrestricted access to critical operations

Technical Details

Affected Platforms:

• ❌ Claude.ai web interface - Permission dialogs completely missing
• ❌ Claude Desktop - Similar permission system failures reported
• ✅ Claude Code terminal - Permission system works correctly
• ✅ Claude.ai mobile/phone - WORKS

Expected Behavior:

1. Claude suggests MCP tool execution
2. Permission dialog appears with Allow/Deny options
3. User explicitly approves or rejects the action
4. Tool executes only after user approval

Actual Behavior:

1. Claude suggests MCP tool execution
2. Tool executes immediately without any permission prompt
3. No user control or approval mechanism visible

Business Impact

This is not just a UX issue - it's a fundamental trust and safety problem:

• Enterprise Adoption Blocked: Companies cannot deploy MCP tools that appear to have unrestricted AI control
• Demo Failures: Cannot demonstrate the intended "AI assistant with human oversight" model
• User Safety Concerns: Users have no way to prevent potentially costly or destructive operations
• MCP Ecosystem Damage: Developers are losing confidence in MCP as a reliable platform

Reproduction Steps

1. Set up any MCP server with tools that include confirmation parameters
2. Connect to Claude.ai web interface
3. Request Claude to perform an action that should require user approval
4. Observe: No permission dialog appears, action executes automatically

Urgency Request

This issue is affecting the broader MCP developer community and blocking production deployments. The permission system is a core feature that enables safe AI-human collaboration. Without it, MCP tools become unusable for any business-critical applications.

Request: Please prioritize this as a P0 issue affecting the entire MCP ecosystem, not just individual users.