CLI-Anything

Generates command-line interfaces for any software, making them controllable by AI agents without relying on fragile UI automation or limited APIs.

Generates command-line interfaces for any software, making them controllable by AI agents without relying on fragile UI automation or limited APIs.

The gist

CLI-Anything is an open-source framework from HKUDS for making software "agent-native." It addresses the challenge of AI agents being unable to use professional software by automatically generating a full-featured command-line interface (CLI) for any given codebase. This allows agents to control complex applications like Blender, GIMP, or QGIS programmatically, bypassing the need for fragile UI automation or limited official APIs, making all software accessible to AI.

What it does

  • Analyzes a software's source code to map its GUI actions and capabilities to APIs.
  • Generates a complete, stateful Python Click CLI with JSON output, a REPL, and undo/redo functionality.
  • Creates a comprehensive test suite with both unit and end-to-end tests for the new CLI.
  • Packages the generated interface into a standard, pip-installable package for easy distribution.
  • Provides a CLI-Hub for agents to autonomously discover, install, and manage available CLIs.
  • Generates SKILL.md documentation files to make generated CLIs discoverable by agent frameworks.

How it works

A user provides a path to a software's source code, often via a Claude Code plugin. CLI-Anything then runs a fully automated 7-phase pipeline that analyzes the code, designs a command structure, implements a Python CLI, generates tests, documents usage, and packages it. The output is a pip-installable package that provides a structured, agent-friendly CLI for the target software. The framework is open-source and free to use.

Best for

Developers and AI engineers who need to automate complex, professional software by making it programmatically controllable for AI agents, especially for applications that lack a robust API.

Watch out for

The primary documented workflow is tightly integrated with specific AI coding agents like Claude Code, which may create a steeper learning curve for users on other platforms. The effectiveness of the generated CLI is dependent on the structure of the target software's codebase.