AI Coreutils — JSON-first CLI for Agents

What is AI Coreutils — JSON-first CLI for Agents?

AI Coreutils is a JSON-first command-line toolkit prototype designed for LLM agents. It is inspired by common GNU Coreutils commands but is not a complete GNU-compatible replacement. It provides a deterministic, low-noise, and easily parsable CLI interface for machine callers, with JSON output by default, JSON errors on stderr, and stable semantic exit codes.

How to use AI Coreutils — JSON-first CLI for Agents?

Install the Python package with pip install aicoreutils. Run commands like aicoreutils schema --pretty, aicoreutils ls . --limit 20, or aicoreutils rm build --recursive --dry-run. For MCP integration, configure a Claude Desktop or AI IDE (Cursor, Windsurf, Continue.dev) to call python -m aicoreutils.mcp_server. Output OpenAI Function Calling format with aicoreutils tool-list --format openai.

Key features of AI Coreutils — JSON-first CLI for Agents

• JSON output by default; JSON errors on stderr.
• Stable semantic exit codes.
• --dry-run for all mutation commands.
• Explicit --raw for pipeline composition.
• 114 CLI commands including agent-native meta-commands.
• MCP server for seamless LLM integration.
• Security profiles: readonly or workspace-write.

Use cases of AI Coreutils — JSON-first CLI for Agents

• Let an LLM agent list files, count lines, or delete directories safely.
• Perform deterministic file operations from within Claude Desktop or Cursor.
• Audit filesystem changes with --dry-run and risk metadata.
• Compose agent pipelines using JSON output and --raw.

FAQ from AI Coreutils — JSON-first CLI for Agents

What is AI Coreutils and how does it differ from GNU Coreutils?

AI Coreutils is a JSON-first CLI for LLM agents inspired by GNU Coreutils. It is not a full GNU clone—it is an agent-friendly subset that prioritizes machine-parseable output, stability, and safety features like --dry-run.

What are the runtime requirements?

Python installed (any platform). The package is available via PyPI (pip install aicoreutils). Tests run with pytest and Hypothesis.

Where does data live, and what security is provided?

Commands operate on the local filesystem. Production security profiles are available: --profile readonly for read-only tools, --profile workspace-write for low-risk writes inside the current working directory.

How is stability guaranteed?

SemVer from v1.0.0. Patch releases fix bugs without changing JSON output. Minor releases add commands/parameters while keeping existing JSON output backward compatible. Major releases may break JSON schema or tool names. LTS version 1.2.3 has critical fixes backported for at least 12 months.

What transports and authentication are supported?

MCP (Model Context Protocol) via the aicoreutils.mcp_server module. Authentication is not described; the server runs with the user's system permissions. OpenAI Function Calling format is also available as an alternative interface.