CLI-Anything
Making ALL Software
Agent-Native
Transform any software into agent-ready tools.
One command to generate CLIs for any codebase.
Quickstart
Open source. One command generates complete CLI for any codebase. Integrates with Claude Code, OpenCode, and Codex. Transform any software into agent-native tools.
7-phase automated pipeline
from code to CLI.
Analyze the codebase.
Scan source code, map GUI actions to APIs, understand software architecture.
Design and implement.
Architect command groups, build Click CLI with REPL, JSON output, undo/redo support.
Test and publish.
Create comprehensive test suite, write documentation, install to PATH.
Everything you need to make
software agent-native.
One-Command Generation
Run /cli-anything on any codebase. Complete CLI with tests, docs, and REPL in one shot.
11 Apps Validated
GIMP, Blender, Inkscape, Audacity, LibreOffice, OBS, Kdenlive, Shotcut, Zoom, Draw.io, AnyGen.
1,508+ Tests
100% pass rate. Unit tests with synthetic data, E2E tests with real software, CLI subprocess verification.
Real Backend Integration
No toy implementations. LibreOffice generates PDFs, Blender renders 3D, actual software via CLI.
Multi-Platform Support
Works with Claude Code, OpenCode, and Codex. Platform-agnostic architecture.
JSON + Human Output
Built-in --json flag for agents. Human-readable tables for debugging. Best of both worlds.
Refine Command
Iteratively expand coverage. Gap analysis finds missing capabilities and implements new commands.
Stateful REPL
Interactive mode with persistent project state, undo/redo, command history, progress indicators.
Zero-Config Install
pip install -e . puts CLI on PATH. Agents discover via standard which commands.
You control the pipeline.
Analyze. Design. Implement. Test. Document. Publish.
CLI-Anything's 7-phase pipeline handles everything automatically. From analyzing the source code to publishing the CLI to PyPI. You get a production-ready interface with comprehensive tests, documentation, and REPL — all in one command.
Build. Refine. Test. Validate. Extend.
After the initial build, use /refine to expand coverage. The gap analysis identifies missing capabilities and implements new commands without breaking existing functionality.
What CLI-Anything is.
Universal CLI generation.
Transform any software with a codebase into agent-ready tools. One command generates complete CLI interfaces.
Authentic software integration.
Direct calls to real applications for actual rendering. No toy implementations — full professional capabilities.
Production-grade testing.
1,508+ tests across 11 major applications. Unit tests, E2E tests, and CLI subprocess verification.
Multi-platform support.
Works with Claude Code, OpenCode, and Codex. Integrates seamlessly with your existing AI workflow.
Agent-native design.
JSON output for machine consumption, human-readable formats for debugging. Agents discover via --help commands.
What changes with CLI-Anything.
AI agents can't use real professional software — only fragile UI automation or limited APIs.
CLI-Anything generates complete CLIs that call real software backends directly. Full capabilities, zero compromises.
Building agent interfaces for software requires manual work, custom integrations, and constant maintenance.
One command generates complete CLI with tests, docs, and REPL. Automated 7-phase pipeline handles everything.
Screenshot-based agents break when UI changes, can't handle complex workflows, and waste tokens on visual processing.
CLI-Anything uses structured commands and JSON output. No screenshots, no clicking, pure command-line reliability.
Agents struggle with scattered APIs, missing context, and incomplete documentation when trying to use software.
Generated CLIs provide --help self-documentation and structured interfaces. Agents discover and use tools naturally.
Custom integrations are expensive, time-consuming, and only cover a fraction of the software's capabilities.
CLI-Anything analyzed entire codebases and generates comprehensive coverage. 1,508 tests across 11 applications validate this.
You need different tools for different platforms — Claude Code, OpenCode, Cursor — each with its own workflow.
CLI-Anything works across multiple platforms with consistent output. One CLI, many agent frameworks.
Why CLI-Anything is special.
CLI-Anything handles the hard CLI generation details correctly.
7-Phase automated pipeline.
Analyze, design, implement, plan tests, write tests, document, and publish — all automated.
Authentic backends.
Real software rendering — LibreOffice generates PDFs, Blender renders 3D, no toy implementations.
JSON + human output.
Built-in --json flag for machine consumption. Human-readable tables for debugging. Best of both.
Refine command.
Iteratively expand coverage with gap analysis. Add missing capabilities without breaking existing ones.
Stateful REPL.
Interactive mode with persistent project state, undo/redo, command history, progress indicators.
Zero-config install.
pip install -e . puts CLI on PATH. Agents discover tools via standard which commands.
Production-grade tests.
1,508 tests with 100% pass rate. Unit, E2E, and subprocess validation across 11 major applications.
What CLI-Anything is not.
Not a GUI automation tool.
No screenshots, no clicking coordinates. CLI-Anything generates structured command-line interfaces.
Not an API wrapper.
We analyze source code and generate authentic CLI integration with real software backends.
Not limited to specific apps.
Works with any software that has a codebase. From image editors to office suites to AI platforms.
Not just for developers.
Once generated, anyone can use the CLI. Agents discover tools via standard --help commands.
Not a toy implementation.
1,508 tests validate real functionality. GIMP edits images, Blender renders 3D, LibreOffice creates documents.
Frequently asked questions.
How is CLI-Anything different from other agent tools?
CLI-Anything transforms software itself into agent-ready tools. Instead of wrapping APIs or automating UIs, it generates authentic CLI interfaces that call real software backends directly.
What platforms does CLI-Anything support?
CLI-Anything works with Claude Code (via plugin marketplace), OpenCode (via slash commands), and Codex (via bundled skill). More platforms coming soon.
What software can CLI-Anything handle?
Any software with a codebase — from image editors (GIMP, Blender) to office suites (LibreOffice) to AI platforms (Stable Diffusion). 11 applications validated with 1,508 tests.
How does the 7-phase pipeline work?
Analyze (scan code), Design (architect CLI), Implement (build Click CLI), Plan Tests (create TEST.md), Write Tests (implement suite), Document (update docs), Publish (create setup.py). All automated.
What is the refine command?
/cli-anything:refine performs gap analysis between the software’s full capabilities and current CLI coverage, then implements new commands and tests for missing functionality.
Do I need the source code?
Yes, CLI-Anything analyzes source code to generate authentic CLIs. For closed-source software, contributions welcome — the methodology is documented in HARNESS.md.
What does a typical setup look like?
Add the marketplace in Claude Code (/plugin marketplace add HKUDS/CLI-Anything), install the plugin (/plugin install cli-anything), then run /cli-anything <software-path>. That’s it.
Can I extend the generated CLI?
Yes. Each CLI is a full Python package you can modify. Use /refine to expand coverage automatically, or edit the code directly for custom changes.
Is CLI-Anything open source?
Yes. MIT licensed. No account required. The methodology, plugin, and all 11 example harnesses are fully open source on GitHub.
From codebase to agent-ready CLI
in one command.
Open source. Works with Claude Code, OpenCode, and Codex. Automatically generates complete CLI with tests, documentation, and REPL. One command to agent-native.