Open Source · Launching 2026

The App Store
for AI Agents

Every CLI tool is a superpower your agent can wield. Discover, install, and create tools with --help as the universal protocol.

$ pip install clihub Browse 500+ Tools →
~/projects/my-app
# Agent discovers tools by searching the store
$ clihub search "resize images batch"
3 tools found · sorted by relevance
image-resize Resize, crop & convert images ★ 4.8 12k installs
sharp-cli High-perf image processing ★ 4.6 8.2k installs
imgp Fast batch image resizer ★ 4.5 5.1k installs
 
# Install in one command, agent-ready immediately
$ clihub install image-resize
Installed image-resize@2.1.0 via pip
Permissions: local-write, network-read
 
# The protocol is --help. That's it.
$ image-resize --help
The Thesis

Why --help beats JSON schemas

MCP promised to be the universal connector. In practice, it eats your agent's context alive. CLI tools are the interface LLMs were born to use.

Token Efficient

MCP dumps full schemas into every prompt. One server with 20 tools adds 8,000–15,000 tokens. CLI tools load nothing until the agent runs --help.

MCP: 72% context wasted CLI: 0% ambient cost
🧠

LLMs Are CLI-Native

Models trained on billions of lines of shell scripts, man pages, and Stack Overflow. They already know git, docker, curl, jq without any schema.

Zero learning cost Training data = docs
🔗

Unix Composability

Pipe stdout to stdin. Chain small tools into powerful workflows. Agents naturally compose commands just like senior engineers write shell scripts.

📖

Self-Documenting

Every CLI ships with --help. Hierarchical discovery: tool --help → resources, tool user --help → actions. No separate schema to maintain.

🔄

Deterministic Feedback

Exit code 0 = success. Exit code >0 = error. Stderr explains what went wrong. The perfect self-healing loop for autonomous agents.

🏗️

Battle-Tested Ecosystem

Thousands of production-grade CLI tools already exist. AWS CLI, GitHub CLI, kubectl, terraform — maintained by the service providers themselves.

For Agents & Humans

Discover. Install. Create. Ship.

Both AI agents and humans are first-class citizens. Use the CLI or the web — same store, same tools.

01

Search with natural language

Semantic search over 500+ indexed tools. Agents search from the terminal, humans from the web. Same results, same ranking.

# Agent searches from terminal $ clihub search "convert PDF to markdown" # Or with flags for structured output $ clihub search "batch audio processing" --json
02

Install in one command

Auto-detects package manager (pip, npm, brew, cargo, docker). Permission tiers enforced. Immediately available on PATH.

$ clihub install pdf-to-md Installed pdf-to-md@1.4.0 via pip Permissions: local-read, local-write $ pdf-to-md --help Convert PDF files to clean Markdown...
03

Create tools with AI

Describe what you want in natural language. The agent generates the CLI tool with --help, --json, tests, and a manifest. Like creating Claude Code skills.

$ clihub create "A tool that extracts tables from PDFs and outputs CSV" Generating CLI with Click... Adding --help documentation... Adding --json output flag... Running automated tests... Created pdf-table-extract in ./pdf-table-extract/
04

Publish to the store

One command to publish. Auto-validated (--help works, examples run, exit codes correct). Community ratings and verified badges.

$ clihub publish ./pdf-table-extract Validating tool.yaml manifest... Testing --help output... Running example commands... Security scan... Published pdf-table-extract@1.0.0 → https://clihub.dev/tools/pdf-table-extract

Simple manifest. Self-documenting CLI.
That's the whole protocol.

No JSON-RPC. No schema injection. No server to maintain. Just a YAML manifest and a CLI that speaks --help.

tool.yaml — Manifest
name: "image-resize" version: "2.1.0" description: "Resize, crop & convert images" author: "@ziyang" license: "MIT" categories: [image, media, conversion] install: method: "pip" package: "image-resize-cli" permissions: [local-read, local-write] agent_hints: when_to_use: "resize or convert images" example: "image-resize in.png -w 800"
SKILL.md — Agent Context (Optional)
--- name: image-resize description: Batch resize and convert images --- ## Usage Resize images with smart cropping. Supports PNG, JPG, WebP, GIF. ## Commands `image-resize <input> -w <px> -o <out>` `image-resize batch <dir> --max-width 1200` `image-resize --json <input>` # structured output ## Tips Use `--json` for agent consumption. Use `--quiet` for piping to other tools.
Marketplace Preview

500+ tools. Growing daily.

Seeded from the best of awesome-cli-apps, plus AI-generated tools from the community. Every tool tested, rated, and agent-ready.

jq Verified
Lightweight command-line JSON processor. Slice, filter, map, and transform structured data.
★ 4.9 ↓ 89k v1.7.1
json data transform
ffmpeg-cli Verified
Convert, process, and stream audio and video. The Swiss Army knife of multimedia.
★ 4.8 ↓ 67k v6.1
video audio media
ripgrep Verified
Blazingly fast recursive search tool. Respects .gitignore. Replaces grep for codebases.
★ 4.9 ↓ 52k v14.1
search code rust
pandoc Verified
Universal document converter. Markdown, LaTeX, HTML, DOCX, PDF, and 40+ formats.
★ 4.7 ↓ 41k v3.5
documents convert markdown
httpie Verified
Human-friendly HTTP client. Colorized output, JSON support, sessions, and plugins.
★ 4.8 ↓ 38k v3.2
http api testing
pdf-table-extract AI-Created
Extract tables from PDF files and output as CSV, JSON, or Markdown. Smart column detection.
★ 4.3 ↓ 2.1k v1.0.0
pdf tables csv ai-created

Give your agent
every tool ever built

One install. 500+ CLI tools. Self-documenting. Agent-native. The Unix way.

$ pip install clihub click to copy
Star on GitHub ★ Read the Docs →