CLIs
47 CLIs available in the registry
Loading CLI registry...
The CLI Registry for AI Agents
If it has an API,
it should have a CLI.
The open-source registry of standardized, agent-ready CLIs. One install. Every API. Any agent.
Paste this into your coding agent.
🦀
🦀OpenClaw Integration
Add any tool to
your OpenClaw
in 2 seconds.
Connect to any API on the planet. Transform it into an agent-ready CLI. Make your OpenClaw so powerful it scares your competitors.
200+ APIs supported
~5KB per CLI
terminal — zsh
$ api2cli create stripe
✓ Scaffolded stripe-cli with 47 resources
$ api2cli bundle stripe
$ api2cli link stripe
Three steps.
That's literally it.
01
Tell your agent
Just say: "Create a CLI for the Stripe API." Your agent uses the api2cli skill to scaffold everything.
> "Create a CLI for the Stripe API"
02
Agent builds it
api2cli reads the API docs, generates typed resources, authentication, error handling. All standardized.
✓ 47 resources · 142 commands · typed
03
Use it everywhere
The CLI works in any terminal, any agent, any IDE. It's just a binary. No setup, no config, no drama.
$ stripe-cli list invoices --status=paid
MCP is dead.Long live CLI.
CLI beats MCP.
Not even close.
MCP was a nice experiment. But experiments end. Production systems need reliability, speed, and universality. CLI delivers all three.
Why MCP fails you
💀
💀
Connection Nightmares
WebSocket drops, stdio crashes, reconnection loops. Your agent stalls while MCP plays dead.
🐌
🐌
Startup Tax: ~800ms
Every. Single. Call. Cold starts, handshakes, capability negotiation. Your agent waits. Your users leave.
🔒
🔒
Vendor Prison
Tied to one IDE. Tied to one runtime. Tied to one protocol version. No escape.
🤮
🤮
Token Obesity
Tool schemas, capability lists, protocol overhead. MCPs eat tokens like they're free. They're not.
VS
Why CLI wins
~25ms
Startup
Native binary speed
0
Dependencies
Single file, self-contained
∞
Compatibility
Works everywhere, forever
100%
Reliability
No connections to drop
| Feature | MCP | CLI |
|---|---|---|
| Startup time | ~800ms | ~25ms |
| Connection | WebSocket/stdio | None needed |
| Dependencies | Runtime + SDK | Zero |
| Works offline | Nope | Always |
| IDE lock-in | Yes | Never |
| Token overhead | ~2K per call | ~200 |
| Debugging | Good luck | Just read stdout |
| Agent discovery | Manual config | Natural |
⚡The paradigm shift
CLI is the
new Skill.
Agent Skills were a breakthrough. But they have limits. CLIs inherit everything skills do — and fix everything they don't.
🧠
Natural Discovery
Models are pre-trained on CLI patterns. They know how to find, invoke, and chain CLIs without instruction. Zero-shot, out of the box.
🪶
Token Efficient
Skills dump entire instruction sets into context. CLIs use --help. 50x less tokens per interaction. Your wallet will thank you.
🔌
No Connection Issues
Skills need MCP bridges, SDK wrappers, or API keys baked in. CLIs are self-contained executables. Run anywhere. Break never.
🔗
Composable by Nature
Pipe output. Chain commands. Parse with jq. CLIs speak Unix. Every tool ever built is already compatible.
Old way: Agent Skill
# SKILL.md — 847 lines # Loaded into context every call # ~12,000 tokens consumed ## Setup Instructions 1. Install dependencies... 2. Configure API keys... 3. Set environment vars... 4. Hope MCP connects...
12K tokens/callFragile
New way: CLI
$ stripe-cli --help Usage: stripe-cli <command> Commands: list List resources get Get a resource create Create a resource $ stripe-cli list customers # Just works. 200 tokens.
~200 tokensBulletproofUniversal
60x
fewer tokens per interaction
Your CFO will send you flowers.
Works with everything.
If it has a terminal, it runs your CLI. No plugins. No extensions. No configuration files. Just works.
Claude Code
Cursor
Windsurf
VS Code
Cline
Aider
Continue
Copilot
Warp
iTerm
Zed
JetBrains
Neovim
Emacs
Any Terminal
15+agent platforms
and counting
and counting
Sponsors