What problem does it solve?
This skill prevents main-agent context pollution by discovering, analyzing, and executing Model Context Protocol (MCP) tool, prompt, and resource capabilities in an isolated subagent environment so LLMs can orchestrate external tools reliably.
Core Features & Use Cases
- Multi-server orchestration: Load and manage multiple MCP servers from a single configuration file to aggregate tool, prompt, and resource catalogs.
- Intelligent tool discovery & selection: Persist discovered capabilities to assets/tools.json for LLM-driven analysis and context-efficient tool selection.
- Execution engine & fallbacks: Run tools via Gemini CLI when available, fall back to direct Node scripts and the mcp-manager subagent; includes lifecycle management, timeouts, and graceful cleanup.
- Use case: Automatically discover web-search, browser automation, and memory tools across servers, select the best tool for a research task, execute it, and save structured results.
Quick Start
Use the mcp-management skill to list all tools across configured MCP servers and save the catalog to assets/tools.json.