MCP Management Skill

Description

This skill manages Model Context Protocol (MCP) servers dynamically, loading them only when needed to avoid context bloat in the main agent. By deferring MCP initialization to a subagent, the main context remains clean even when multiple MCP servers are configured.

Purpose

• •Load MCP servers from .claude/.mcp.json on-demand
• •List available MCP tools without polluting main context
• •Execute MCP tool calls and return results
• •Keep main agent context pristine and efficient

Key Benefits

• •Context Efficiency: MCP tools only consume subagent context, not main context
• •Scalability: Handle multiple MCP servers without context bloat
• •Flexibility: Dynamically load only the MCP servers you need
• •Performance: Main agent remains fast and responsive

How It Works

1. •Configuration: MCP servers are configured in .claude/.mcp.json (not the default location)
2. •On-Demand Loading: When you need MCP tools, invoke the mcp-manager subagent
3. •Tool Discovery: Subagent loads MCP servers and lists available tools
4. •Tool Execution: Subagent executes the requested tool and returns results
5. •Context Isolation: All MCP overhead stays in the subagent's context

Usage

Step 1: Configure MCP Servers

Create or move your MCP configuration to .claude/.mcp.json:

{
  "mcpServers": {
    "chrome-devtools": {
      "command": "npx",
      "args": ["-y", "chrome-devtools-mcp"]
    },
    "playwright": {
      "command": "npx",
      "args": ["-y", "@playwright/mcp-server"]
    }
  }
}

Step 2: Use via mcp-manager Subagent

Instead of using MCP tools directly, invoke the mcp-manager subagent:

1. •"I need to take a screenshot of a website"
2. •mcp-manager subagent loads MCP servers
3. •Discovers chrome-devtools or playwright tools
4. •Executes screenshot tool
5. •Returns result to main agent

Scripts Available

• •list-mcp-servers.sh: List all configured MCP servers
• •list-mcp-tools.sh: List all available tools from configured servers
• •check-mcp-config.sh: Validate MCP configuration file

Notes

• •This approach trades token efficiency for execution speed (subagent invocation overhead)
• •Best used when you have many MCP servers but only need them occasionally
• •For frequently used MCP tools, traditional direct usage may be more efficient
• •The subagent handles all MCP complexity, keeping main agent focused

Comparison: Traditional vs Subagent Approach

Traditional Approach

• •✅ Fast tool execution (direct)
• •❌ All MCP tools loaded into main context from start
• •❌ Context bloat with multiple servers
• •❌ Token inefficiency

Subagent Approach

• •✅ Main context stays clean
• •✅ Scale to many MCP servers
• •✅ Token efficient
• •❌ Slight overhead from subagent invocation
• •❌ Extra step in workflow

Credits

Inspired by @goon_nguyen's approach to MCP context management: https://x.com/goon_nguyen/status/1987720058504982561