Instead of maintaining a catalog of tools like read_file, grep, list_dir, they just expose one run(command: str) tool and let the LLM compose Unix pipelines. It has access to one tool for shell commands: run(...)
I think you could use mcp-shell , written in Go instead of python subprocesses
He discusses:
Instead of maintaining a catalog of tools like read_file, grep, list_dir, they just expose one run(command: str) tool and let the LLM compose Unix pipelines. It has access to one tool for shell commands: run(...)
I think you could use mcp-shell , written in Go instead of python subprocesses