[Bug] Opus no longer performs validation checks

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/6894
2. https://github.com/anthropics/claude-code/issues/6401
3. https://github.com/anthropics/claude-code/issues/5541

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

Oh the lying is another big thing and always saying, "You are absolutely right!. It starts well but the moment you correct it, it goess off completely.

Feature Analysis: MCP Enable/Disable CLI Commands

Excellent, detailed feature request with clear use cases. This would significantly improve context management.

I Already Proposed This in My Plugin PR!

See: https://github.com/anthropics/claude-code/pull/12356

The quick-wins plugin includes a pattern for hook-based MCP management, but native CLI support is better.

Proposed Implementation

Add two new commands:

claude mcp enable <server-name>
claude mcp disable <server-name>
claude mcp status [server-name]

Implementation:

// src/cli/commands/mcp.ts

interface MCPCommand {
  action: 'enable' | 'disable' | 'status';
  serverName?: string;
}

async function handleMCPCommand(cmd: MCPCommand) {
  const session = getCurrentSession();
  
  switch (cmd.action) {
    case 'enable':
      await session.mcp.enable(cmd.serverName);
      console.log(`✓ Enabled ${cmd.serverName}`);
      console.log(`Context freed: ${calculateContextSaved(cmd.serverName)}K tokens`);
      break;
      
    case 'disable':
      await session.mcp.disable(cmd.serverName);
      console.log(`✓ Disabled ${cmd.serverName}`);
      break;
      
    case 'status':
      const servers = await session.mcp.list();
      servers.forEach(s => {
        console.log(`${s.enabled ? '●' : '○'} ${s.name} (${s.tokenUsage}K tokens)`);
      });
      break;
  }
}

Session-level state management:

// src/mcp/SessionMCPManager.ts

class SessionMCPManager {
  private enabledServers: Set<string>;
  
  async enable(serverName: string) {
    // Load server config
    const config = this.loadServerConfig(serverName);
    
    // Initialize MCP client
    const client = new MCPClient(config);
    await client.connect();
    
    // Register tools in session
    this.session.registerTools(await client.listTools());
    
    this.enabledServers.add(serverName);
  }
  
  async disable(serverName: string) {
    // Unregister tools
    const client = this.clients.get(serverName);
    const tools = await client.listTools();
    this.session.unregisterTools(tools);
    
    // Disconnect
    await client.disconnect();
    
    this.enabledServers.delete(serverName);
  }
}

Hook Integration

PreToolUse hook:

# .claude/hooks/pre-tool-use.sh
#!/bin/bash

TOOL_NAME=$1

# Enable Playwright only when browser tools are used
if [[ "$TOOL_NAME" == "browser_navigate" || "$TOOL_NAME" == "browser_click" ]]; then
  claude mcp enable playwright
fi

SubagentStop hook:

# .claude/hooks/subagent-stop.sh
#!/bin/bash

SUBAGENT_NAME=$1

if [[ "$SUBAGENT_NAME" == "accessibility-compliance-expert" ]]; then
  claude mcp disable playwright
  echo "♻️ Freed context: disabled Playwright MCP"
fi

Configuration File

// .claude/mcp.json
{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-playwright"],
      "autoEnable": false,  // Don't enable at session start
      "enabledBy": ["hook:pre-tool-use"]  // Enable via hook
    }
  }
}

Context Optimization Logic

Auto-disable idle servers:

class IdleServerMonitor {
  private lastUsed: Map<string, Date>;
  
  startMonitoring() {
    setInterval(() => {
      const now = new Date();
      
      for (const [server, lastUse] of this.lastUsed) {
        const idleMinutes = (now.getTime() - lastUse.getTime()) / 60000;
        
        if (idleMinutes > 30) {  // 30 min idle
          this.session.mcp.disable(server);
          console.log(`♻️ Auto-disabled ${server} (idle 30min)`);
        }
      }
    }, 60000);  // Check every minute
  }
}

UI Integration

Status in prompt:

> /config

MCP Servers:
  ● pieces-mcp (5.2K tokens)
  ○ playwright (15.3K tokens) [DISABLED]
  ● sequential-thinking (2.1K tokens)

Total context used: 7.3K / 200K (3.7%)

Security Considerations

Respect managed-mcp.json:

async function enable(serverName: string) {
  const managed = await loadManagedConfig();
  
  if (managed.blockedServers.includes(serverName)) {
    throw new Error(`Server ${serverName} is blocked by enterprise policy`);
  }
  
  // Continue with enable...
}

Alternative: /mcp Slash Command

For interactive use:

> /mcp enable playwright
✓ Enabled playwright (context +15K tokens)

> /mcp disable playwright  
✓ Disabled playwright (context -15K tokens)

> /mcp status
● pieces-mcp (5.2K)
○ playwright (15.3K) [DISABLED]

Implementation Estimate

Week 1: CLI commands + session-level enable/disable
Week 2: Hook integration + context monitoring
Week 3: Auto-idle detection + UI polish

---

Verdict: HIGH value feature, enables powerful automation workflows.
Recommend: Prioritize for next release (v2.1.0).

For now, see my workaround in https://github.com/anthropics/claude-code/pull/12356

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

Still occurring

This issue has been automatically closed due to 60 days of inactivity. If you're still experiencing this issue, please open a new issue with updated information.

This issue was closed incorrectly despite recent human comments. This behavior of the bot is reported at https://github.com/anthropics/claude-code/issues/16497. Please upvote that issue, so maybe it gets noticed.

More reasons to switch to Codex.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.