mcp-management

To manage multiple MCP servers without polluting the context window, use a skill that applies progressive disclosure to load only necessary tool definitions and delegates operations to a subagent. This keeps your main context clean while orchestrating servers.

Automated tool discovery for the Model Context Protocol is the process of analyzing configured servers to identify and catalog relevant tools, prompts, and resources for specific tasks. It maintains a persistent catalog for fast reference and context-efficient execution.

To configure multiple MCP servers for CLI operations, you need to define your server connections within a .claude/.mcp.json config file. Once configured, you can run the management CLI to list available tools and call them with proper parameter handling.

Yes, MCP management works directly with the @modelcontextprotocol/sdk, which serves as its required dependency. This integration enables the execution engine to call MCP tools and handle parameters correctly across multiple configured servers.

The best way to execute tools across different MCP servers is to use an orchestration skill that applies intelligent discovery to analyze task relevance. It connects to your servers via a single config and handles execution without overloading the main context.

Your LLM context window overflows when loading MCP tools because all tool definitions are loaded simultaneously instead of using progressive disclosure. You need a management skill that maintains a persistent catalog and loads only necessary definitions for context efficiency.