Name: CliHub
Author: CliHub

~/projects/my-app

# Agent discovers tools by searching the registry

$ clihub search "resize images batch"

2 tools found · sorted by relevance

● imagemagick Create, edit, compose digital images ★ 4.7 58k

● ffmpeg The Swiss Army knife of multimedia ★ 4.8 67k

# One command to install, agent-ready immediately

$ clihub install imagemagick

✓ Installed imagemagick (v7.1.1) via brew

ℹ Permissions: local-read, local-write

# The protocol is --help. That's it.

$ magick --help

Usage: magick [options] input output

The Paradigm Shift

The user is no longer human.

GUI apps were built for people who click buttons. AI agents don't have eyes or hands — they have stdin, stdout, and --help.

	🧑 GUI Apps	🔧 MCP	🤖 CLI (--help)
Agent-usable?	No — needs a human	Yes, with schema	Yes, natively
Context cost	N/A	8k–15k tokens / server	0 tokens until --help
Discovery	Manual browsing	Schema in every prompt	On-demand: clihub search
LLM familiarity	Can’t use at all	Needs schema learning	Trained on billions of examples
Composability	Copy-paste between apps	Single-tool calls	Unix pipes: curl \| jq \| csv
Error handling	Dialog boxes	Custom per server	Exit codes + stderr
Automation	Fragile screen scraping	Build & maintain server	Already on PATH
Structured output	Screenshots	JSON-RPC responses	--json on every command

The App Store gave humans an app for everything. CliHub gives agents a tool for everything. GUI apps need screens, clicks, and humans in the loop. CLI tools need nothing but a shell — and LLMs were trained on billions of shell commands. They already know git, curl, jq. No schema. No learning curve. The software paradigm has shifted.

For Agents & Humans

Search. Install. Explore. Go.

Seven commands. That's the entire CLI. Both AI agents and humans are first-class citizens.

clihub --tour

# Search with natural language

$ clihub search "convert PDF to markdown"

1 tool found

● pandoc Universal document converter ★ 4.7 41k

# Agents use --json for structured output

$ clihub search "json processing" --json

[{"name":"jq","rating":4.9,"install":"brew"}, ...]

# Auto-detects the right package manager

$ clihub install pandoc

✓ Installed pandoc (v3.6.1) via brew

ℹ Permissions: local-read, local-write

# Immediately available on PATH

$ pandoc --help

pandoc [OPTIONS] [FILES]

Convert between document formats...

# Get full metadata + --help output for any tool

$ clihub info jq

┌──────────────────── jq ────────────────────┐

│ jq v1.7.1 │

│ Lightweight command-line JSON processor │

│ Install: brew install jq │

│ When to use: Parse, filter, or │

│ transform JSON data │

└───────────────────────────────────────────┘

# Browse the full registry by category

$ clihub list --category data

data (6 tools)

● jq Lightweight JSON processor ★ 4.9

● yq YAML/JSON/XML/CSV processor ★ 4.7

● csvkit CSV toolkit ★ 4.5

● xsv Fast CSV toolkit (Rust) ★ 4.6

● dasel Multi-format data query ★ 4.4

● gron Make JSON greppable ★ 4.5

# Or check what's already on your system

$ clihub list --installed

Search tools with natural language. Fuzzy matches against name, description, and tags. Use --json for structured output.

The Protocol

The protocol is `--help`.
That's it.

No JSON-RPC. No schema injection. No server to maintain. A bundled registry, --json on every command, and exit codes that work.

{
  "name": "jq",
  "version": "1.7.1",
  "description": "Lightweight command-line JSON processor",
  "author": "Stephen Dolan",
  "homepage": "https://jqlang.github.io/jq/",
  "categories": ["data", "json"],
  "install": {
    "method": "brew",
    "package": "jq",
    "binary_name": "jq"
  },
  "agent_hints": {
    "when_to_use": "Parse, filter, or transform JSON data",
    "example_usage": [
      "cat data.json | jq '.users[]'",
      "jq -r '.items[].name' file.json"
    ]
  },
  "rating": 4.9,
  "verified": true
}

$ clihub --json search "json processing"

[
  {
    "name": "jq",
    "version": "1.7.1",
    "description": "Lightweight command-line JSON processor...",
    "install_method": "brew",
    "rating": 4.9,
    "installed": true
  },
  {
    "name": "gron",
    "version": "0.7.1",
    "description": "Make JSON greppable...",
    "install_method": "brew",
    "rating": 4.5,
    "installed": false
  }
]

# Exit code 0 = results found
# Exit code 2 = no results

Marketplace

100+ tools. All agent-ready.

Battle-tested CLI tools from the open-source ecosystem. Every tool verified, rated, and self-documenting via --help.

Add Your Tool

Any CLI tool. Three commands.

Already have a CLI tool on your system? Convert it to a CliHub registry entry automatically.

Convert — auto-detect metadata

Point clihub convert at any binary on your PATH. It detects version, description, author, license, and install method automatically.

$ clihub convert jq
✓ Detected jq v1.7.1 (brew)
✓ Generated clihub.yaml

Edit — fill in agent hints

Open clihub.yaml and fill in the TODOs: categories, tags, and agent_hints so AI agents know when and how to use your tool.

categories: ["data", "json"]
tags: ["json", "filter", "transform"]
agent_hints:
  when_to_use: "Parse, filter, or transform JSON data"
  example_usage:
    - "cat data.json | jq '.users[]'"

Submit — validate and open a PR

Validates your manifest (name format, description length, required fields) and outputs PR-ready JSON. Then open a pull request.

$ clihub submit clihub.yaml
✓ Name format valid
✓ Description: 10-200 chars
✓ Categories provided
✓ No duplicate in registry

→ JSON ready. Open a PR to clihub-ai/clihub

Already have a tool on GitHub? You can also edit src/clihub/data/registry.json directly and submit a PR. Each entry is a JSON object with name, description, install method, and agent_hints. See the contributing guide for details.

Humans got the App Store.
Agents get CliHub.

100+ tools. Zero tokens wasted. Self-documenting. The software user has changed. The store should too.

$ pip install clihub-ai click to copy

or: pipx install clihub-ai · Requires Python 3.10+

Star on GitHub Read the Docs →

The App Storefor AI Agents