MCP Management Skill

Load MCP servers on demand into a subagent and execute tool calls.

1|Updated Feb 3, 2026
One-click install
npx skills add https://github.com/run6270/skill --skill mcp-management-skill
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: MCP Management Skill
Source: https://github.com/run6270/skill/tree/main/mcp-management
Command: npx skills add https://github.com/run6270/skill --skill mcp-management-skill

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

This skill addresses the issue of context bloat in the main agent by dynamically loading Model Context Protocol (MCP) servers only when needed, ensuring a clean and efficient main agent context.

Core Features & Use Cases

  • On-Demand Loading: Initializes MCP servers from a specified configuration file (.claude/.mcp.json) only when required.
  • Context Isolation: MCP server operations are confined to a subagent, preventing them from consuming the main agent's context.
  • Tool Discovery & Execution: Enables listing available MCP tools and executing tool calls through the subagent.
  • Use Case: When you need to interact with a specific MCP tool, like taking a website screenshot, you invoke the mcp-manager subagent, which then loads the necessary MCP server and executes the tool, returning the result without cluttering your main agent's context.

Quick Start

Use the mcp-manager subagent to take a screenshot of a website.

Frequently Asked Questions about MCP Management Skill

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I prevent context bloat when using multiple MCP servers?

To prevent context bloat with multiple MCP servers, you can dynamically load them into a subagent on demand. This isolates server operations and tool execution, keeping the main agent's context clean and efficient.

How do I load MCP servers on demand from a configuration file?

You can load MCP servers on demand by initializing them from a specified configuration file, such as `.claude/.mcp.json`. The skill fetches the server definitions only when required, preventing unnecessary context consumption.

Can I list available MCP tools and execute calls without cluttering the main agent context?

Yes, you can list available MCP tools and execute calls without context clutter by invoking a subagent. The subagent handles the tool discovery and execution, returning only the final result to the main agent.

What is the best way to manage dynamic MCP server loading for tool execution?

The best way to manage dynamic MCP server loading is to use a subagent that handles initialization and tool calls. This approach supports multiple servers without bloating the main context and offers flexible, on-demand execution.

Does MCP server management support taking a website screenshot through a subagent?

Yes, MCP server management supports taking a website screenshot through a subagent. When you need a specific tool, the subagent loads the necessary MCP server, executes the screenshot tool, and returns the result.

Why does loading multiple MCP servers consume the main agent context?

Loading multiple MCP servers consumes context because server operations and tool outputs normally reside in the main agent. Isolating these operations within a subagent prevents the main context from filling up with server data.