What problem does it solve? Interacting with Model Context Protocol servers normally requires loading full tool schemas into an AI agent's context, consuming thousands of tokens, or writing custom client code. This Skill provides a lightweight CLI to discover, inspect, and call MCP tools directly from the shell with JSON output. ## Core Features & Use Cases - On-Demand Schema Loading: List servers, grep tools by glob pattern, and fetch individual tool schemas only when needed, saving context tokens. - Shell Composability: JSON output pipes cleanly into jq for filtering, chaining multiple MCP calls, and building automation scripts. - Flexible Configuration: Supports stdio and HTTP servers, environment variable substitution, tool allow/deny filtering, and connection pooling via a lazy-spawn daemon. - Use Case: An AI agent needs to search GitHub repositories through an MCP server. Instead of loading all schemas, it runs mcp-cli info github search_repositories to get one schema, then calls the tool and pipes the result through jq to extract repository URLs. ## Quick Start Install mcp-cli, create an mcp_servers.json config with your MCP servers, then ask the agent to list available tools and call one with JSON arguments.
