A CLI tool for dynamically managing MCP server configurations in Claude Code. Enable and disable MCP servers on-demand to optimize context usage and performance.

pnpx mcpick
# or
pnpm dlx mcpick
# or
npx mcpick

pnpm install -g mcpick
# or
npm install -g mcpick

Using the Claude Code /doctor command you may see something like this if you have many MCP servers configured:

 Context Usage Warnings
 └ ⚠ Large MCP tools context (~66,687 tokens > 25,000)
   └ MCP servers:
     └ mcp-omnisearch-testing: 20 tools (~10,494 tokens)
     └ mcp-omnisearch: 20 tools (~10,454 tokens)
     └ mcp-sqlite-tools-testing: 19 tools (~9,910 tokens)
     └ mcp-sqlite-tools: 19 tools (~9,872 tokens)
     └ playwright: 21 tools (~9,804 tokens)
     └ (7 more servers)

Claude Code loads all MCP servers from your .claude.json file at startup, regardless of whether you need them for your current task. This can lead to:

• 🐌 Slower Claude Code startup times
• 💾 High context token usage
• 🧠 Cognitive overload from too many available tools

McPick provides an intuitive CLI menu to:

• ✅ Toggle servers on/off - Enable only the MCP servers you need for your current task
• 📁 Manage server registry - Keep a database of all your available MCP servers
• 🔄 Safe configuration - Only modifies the mcpServers section, preserving other Claude Code settings
• 💾 Backup & restore - Create focused backups of your MCP server configurations

┌  McPick - MCP Server Configuration Manager
│
◆  What would you like to do?
│  ● Enable / Disable MCP servers (Toggle MCP servers on/off)
│  ○ Backup config
│  ○ Add MCP server
│  ○ Restore from backup
│  ○ Load profile
│  ○ Save profile
│  ○ Exit
└

MCPick supports the three MCP server scopes used by Claude Code:

When you select "Enable / Disable MCP servers", MCPick will:

1. Ask which scope you want to edit
2. Show servers already enabled for that scope (pre-checked)
3. Use claude mcp add/remove CLI commands for Local and Project scopes

This integration ensures your changes are correctly applied to the right configuration location.

• Auto-discovery: Automatically imports servers from your existing .claude.json
• Registry sync: Maintains a registry of available servers for quick selection
• Selective enabling: Choose exactly which servers to enable via multiselect
• Configuration safety: Preserves all non-MCP settings in your Claude Code config

• Focused backups: Only backs up MCP server configurations (not the entire 30k+ line config)
• Automatic cleanup: Keeps last 10 backups to prevent storage bloat
• Easy restoration: Restore from any previous backup with a simple menu

Load predefined sets of MCP servers instantly:

# Apply a profile
mcpick --profile database
mcpick -p database

# Save current config as a profile
mcpick --save-profile mysetup
mcpick -s mysetup

# List available profiles
mcpick --list-profiles
mcpick -l

Profiles are stored in ~/.claude/mcpick/profiles/. You can also create them manually:

// ~/.claude/mcpick/profiles/database.json
{
	"mcp-sqlite-tools": {
		"type": "stdio",
		"command": "npx",
		"args": ["-y", "mcp-sqlite-tools"]
	}
}

Or use full format with mcpServers wrapper:

{
  "mcpServers": {
    "server-name": { ... }
  }
}

1. Before a coding session: Run mcpick -p <profile> or use the interactive menu to enable relevant servers
2. Launch Claude Code: Run claude to start with your configured servers
3. Switch contexts: Run mcpick -p <other-profile> to quickly switch server sets

1. Select "Add MCP server"
2. Provide server details:
   • Name (e.g., "mcp-sqlite-tools")
   • Command (e.g., "npx")
   • Arguments (e.g., "-y", "mcp-sqlite-tools")
   • Description (optional)
   • Environment variables (optional)

MCPick works with the standard Claude Code configuration format:

{
	"mcpServers": {
		"server-name": {
			"command": "npx",
			"args": ["-y", "mcp-server-package"],
			"env": {
				"API_KEY": "your-key"
			}
		}
	}
}

• Claude Config: ~/.claude.json (your main Claude Code configuration)
• Project Config: .mcp.json (project-specific shared config, committed to git)
• MCPick Registry: ~/.claude/mcpick/servers.json (MCPick's server database)
• Backups: ~/.claude/mcpick/backups/ (MCP configuration backups)
• Profiles: ~/.claude/mcpick/profiles/ (predefined server sets)

Note: MCPick automatically detects servers in parent directories. If you have local servers configured at /Users/you/projects and run MCPick from /Users/you/projects/myapp, it will find and display them.

• Non-destructive: Only modifies the mcpServers section of your Claude Code config
• Backup integration: Automatically creates backups before major changes
• Validation: Ensures all server configurations are valid before writing
• Error handling: Graceful failure modes with helpful error messages

McPick is actively being developed with new features planned. See the roadmap for details on:

• Settings Validation - Validate your Claude Code settings files using the claude-code-settings-schema
• Permissions Management - Interactive tool permission configuration with presets (Safe Mode, Dev Mode, Review Mode)

Have ideas for other features? Open an issue or check out the contribution guide!

• Node.js 22+
• Claude Code installed and configured
• pnpm (for building from source)