Feature Request: Programmatic API for enabling/disabling MCPs without restart

Problem

Currently, users can enable/disable MCP servers in two ways:

1. Via /mcp interactive menu - Works without restart (hot-reload) ✅
2. Via editing .claude.json - Requires Claude Code restart ❌

While the interactive menu is great for manual control, there's no programmatic way to toggle MCPs on/off without restarting, which limits automation capabilities.

Use Case

Users want to create custom skills/commands that can dynamically enable/disable MCPs based on context. For example:

• Project-specific MCP profiles: Enable only relevant MCPs for a project type
• Performance optimization: Disable heavy MCPs during specific tasks
• Batch operations: Toggle multiple MCPs programmatically
• Automated workflows: Enable/disable MCPs as part of a script or hook

Example desired workflow:

# Custom skill that prepares environment for frontend work
/setup-frontend
→ Disables: mysql, postgres, docker MCPs
→ Enables: playwright, browser, figma MCPs
→ No restart needed

Current Investigation

I've investigated the codebase structure and found:

What works:

• .claude.json structure:

``json
{
"projects": {
"/path/to/project": {
"disabledMcpServers": ["context7", "slack"]
}
}
}
``

• /mcp menu successfully modifies this structure AND triggers hot-reload
• Skills can modify .claude.json but changes only apply after restart

What's missing:

• No programmatic API to trigger the internal reload function that /mcp uses
• Skills have no access to the runtime that manages MCP processes

Proposed Solution

Add one of the following:

Option 1: Expose MCP control via Bash tool commands

claude mcp enable context7
claude mcp disable slack
claude mcp reload  # Force reload config

Option 2: Add dedicated MCP control tool

Similar to existing tools (Read, Write, Bash), add:

• MCPEnable(server_name)
• MCPDisable(server_name)
• MCPReload()

Option 3: Add skill-accessible IPC endpoint

Allow skills to send commands to Claude Code runtime:

# Skills could use this internally
echo '{"action":"mcp.disable","server":"context7"}' > ~/.claude/.runtime/commands.fifo

Benefits

1. Automation: Users can create sophisticated workflows
2. Context-aware: MCPs auto-enable/disable based on project
3. Performance: Programmatically disable unused MCPs
4. Developer experience: More powerful customization

Additional Context

• Current behavior: Modifying .claude.json requires restart
• Desired behavior: Programmatic changes apply immediately (like /mcp menu)
• This would unlock powerful automation capabilities while maintaining the same UX as the interactive menu

Related

This would pair well with existing features:

• Hooks (.claude/settings.json)
• Custom commands (.claude/commands/)
• Project-specific configs (.claude.json per project)

---

Would love to see this feature! It would significantly enhance Claude Code's automation and customization capabilities.