CLI-Anything Hub: agent-native CLIs for any software
A generator that turns software into a full CLI with tests and a REPL, plus a public registry where agents and humans install those CLIs by name. How it works, the use cases it is built for, and where it stops.
11 min read
On this page
CLI-Anything starts from a blunt premise: if software has a codebase or an API, an AI agent should be able to drive it from the terminal instead of clicking through a GUI. It does two things. First, a generator that turns a codebase into a full command-line interface in one command, complete with tests and a REPL. Second, CLI-Hub at clianything.cc, a public registry where those generated CLIs and curated public ones are published and installed by name. It leaves the software alone and gives an agent the same knobs a human has, as structured commands with JSON output, so it can do real work without screenshots.
What CLI-Anything is
CLI-Anything is an open-source project (MIT, from HKUDS) that analyzes a software project's source and generates a Click-based Python CLI. The generated tool calls the real backend, not a stub: LibreOffice actually produces PDFs, Blender actually renders 3D, GIMP actually edits images. The project ships eleven validated harnesses (GIMP, Blender, Inkscape, Audacity, LibreOffice, OBS Studio, Kdenlive, Shotcut, Zoom, Draw.io and AnyGen) and over 1,500 tests. It works with Claude Code, OpenCode and Codex.
- Open source under MIT, no account required.
- One command generates the CLI, tests, documentation and a REPL.
- Generated CLIs install with pip install -e . and expose a --json flag.
- A refine command finds gaps and adds commands without breaking existing ones.
- Works across Claude Code, OpenCode and Codex with consistent output.
How the generator works
You point the generator at a codebase and it runs a seven-phase pipeline. The output is a real Python package on your PATH, not a wrapper around screenshots.
- Analyze the codebase: scan source, map GUI actions to APIs, understand the architecture.
- Design: architect command groups and the overall CLI shape.
- Implement: build the Click CLI with a REPL, JSON output and undo/redo.
- Plan tests: write a TEST.md describing the coverage.
- Write tests: unit, end-to-end and subprocess verification.
- Document: update README and help text.
- Publish: package and install to PATH.
CLI-Hub: the registry
CLI-Hub at clianything.cc is the registry side. It hosts generated harness CLIs plus curated public CLIs, and the cli-anything-hub package installs the cli-hub command. From there you browse, search, inspect, install, update, uninstall and launch tools. The hub also ships workflow matrices: capability bundles that map a task such as 'make a captioned video' to several providers, with preflight checks for what is actually usable on your machine.
- cli-hub list: every CLI, grouped by category, with an installed marker.
- cli-hub search <term>: search by name, description or category.
- cli-hub info <name>: full detail for one CLI.
- cli-hub install <name>: install from the registry.
- cli-hub launch <name> [args]: run an installed CLI.
- cli-hub matrix ...: capability bundles, preflight, install by capability, recipes and doctor.
Workflow matrices
A matrix is the part that matters for agents. A real task rarely maps to one tool. A matrix bundles a domain into capabilities, the verbs an agent calls, and providers, the concrete ways to fulfil them. Preflight reports which providers are actually available, so the agent decides what to install before touching anything. Paid APIs are never used without configured keys and your consent. After installing, the matrix renders a SKILL.md with provider-selection rules and references.
Where this actually gets used
The hub's demos show agents producing finished artifacts without a GUI. A few patterns recur, and they are the use cases the tool is built for.
Captioned video from a script
An agent uses a video-creation matrix to transcribe audio, generate or select visuals, edit and export. Providers might include ffmpeg as a native binary, a Python library, or a cloud API; preflight shows which are installed before anything runs. The agent chains capabilities and exports an MP4 without opening an editor.
A CAD model iterated by text
The FreeCAD harness lets an agent build a parametric part by command instead of clicking the workbench. That suits generating variants, checking dimensions, and exporting STEP files in a loop, where repeating the same clicks by hand is the slow part.
A diagram generated from a description
The Draw.io harness builds a full diagram from scratch via CLI. The site shows an HTTPS handshake lifecycle, TCP handshake through TLS negotiation to termination, built entirely through commands. That fits documentation pipelines where a diagram has to match the code it describes.
Batch document conversion
The LibreOffice harness converts documents headlessly with real rendering. An agent can walk a folder, convert files to PDF, and verify outputs. That is tedious by hand and trivial by command, which is the whole pitch.
Scouting a capability before installing
cli-hub can 'transcribe audio' searches all matrices and reports which providers exist and which are missing on this machine. For an agent, that is the useful bit: decide what to install before touching anything, instead of failing halfway through a task.
Quick start
- Install the hub: pip install cli-anything-hub (Python 3.10 or newer).
- Find a tool: cli-hub list or cli-hub search <term>.
- Install it: cli-hub install <name>, then run cli-hub launch <name>.
- For agents: npx skills add HKUDS/CLI-Anything --skill cli-hub-meta-skill -g -y, then prompt 'Find appropriate CLI software in CLI-Hub and complete the task: ...'.
- To generate your own: install the CLI-Anything plugin in Claude Code, OpenCode or Codex and run /cli-anything ./your-app.
When to reach for it, and when not
Reach for CLI-Anything when you want an agent to operate real software (rendering, conversion, editing) instead of UI automation, when you are building a pipeline that mixes several tools and want capability-based discovery, or when you need machine-readable output and exit codes for scripting.
Skip it when the software has no source code and no public API, because the generator needs a codebase and closed-source support is community-contributed; when a web UI is faster for a one-off; or when a task needs a paid API without a key and your consent, since those are never used automatically.
Today's software serves humans. Tomorrow's users will be agents.
CLI-Anything's bet is that the next user of most software is an agent, and the interface should be a command, not a screen. Whether that holds for your stack comes down to whether your tools have a CLI today. If they do not, the generator is the interesting part; if they do, the hub is.
If you found this useful, browse the rest of the YesIntelligent articles for more practical guides on creator tools, open-source software and reading engagement honestly.
Related articles
Every free tool YesIntelligent offers, and when to use each
Nine small tools that cover the repetitive jobs of a creator's week: pulling the words out of a video, checking a post before it gets cut off, reading an engagement number honestly, turning a good line into an image, finding prospects, and rewriting stiff AI copy.
YouTube transcript generator: use cases and why ours
A YouTube transcript is the fastest way to turn a video you admire into text you can search, quote, and reuse. Here are the jobs it does best, and why ours stays out of your way.
Instagram Reel transcript generator: use cases and why ours
Reels are short, fast, and often uncaptioned. A transcript turns each one into text you can study in a minute, even when nothing is written on screen.