Exposes 114 JSON-first coreutils commands (ls, rm, wc, etc.) through MCP for AI agents to manipulate filesystems programmatically. Every command returns structured JSON by default, errors go to stderr as JSON, and mutations support dry-run mode. Drop it into Claude Desktop's config and your assistant can navigate directories, count lines, move files, all with machine-readable responses instead of parsing terminal output. Ships with security profiles for read-only or workspace-scoped writes. Outputs OpenAI function calling schemas via tool-list if you're wiring it into LangChain or custom agents. Built for deterministic CLI operations when you need agents to perform file tasks without brittle regex scraping.
claude mcp add --transport stdio io.github.caseshy-aicoreutils uvx aicoreutils