MCPShell is a tool that allows LLMs to safely execute command-line tools through the Model Context Protocol (MCP). It provides a secure bridge between LLMs and operating system commands. It supports flexible command execution by turning any shell commands into MCP tools, with parameter substitution via templates. Tools are defined in YAML with parameters, constraints, and output formatting. Security is enforced by validating tool parameters using CEL expressions before execution, and optional sandboxed environments for running commands (see docs/config-runners.md). It enables rapid prototyping of MCP tools: you can drop in shell code and reuse it as an MCP tool in your LLM workflows. The server is designed for easy integration with MCP clients, including Cursor, Visual Studio Code, and Wisty, and includes guidance in the Quick Start on configuring the client with a tools directory. The project also offers an Agent Mode that connects MCPShell to an OpenAI-compatible API so models can autonomously invoke defined tools, with system and user prompts configurable in the YAML.