May 27, 2026 · 8 min read

Three Ways to Distribute Your CLI to AI Agents

You built a CLI tool. It works. Developers use it from the terminal. But here's the problem: AI coding agents can't run terminal commands natively. Claude Code, Cursor, Codex, Copilot — they all use the Model Context Protocol (MCP) to call tools. If your CLI isn't an MCP server, it's invisible to every AI agent.

click-to-mcp has always solved this with one command: click-to-mcp serve your-cli. But the AI agent ecosystem has grown fast, and now there are three distinct distribution channels you need to cover. Here's what changed and how to reach all of them.

The Problem: Three Ecosystems, One CLI

The AI agent landscape has consolidated around three distribution mechanisms, each with different install flows, discovery paths, and user expectations:

Most CLI authors stop at MCP. That's a mistake — Claude Code plugins and Agent Skills each reach audiences that don't browse MCP directories. Here's how to cover all three.

Channel 1: MCP Server (The Foundation)

This is what click-to-mcp was built for. Wrap any Click or Typer CLI as an MCP server with a single command:

pip install click-to-mcp
click-to-mcp serve your-package.cli:app

Every CLI command becomes an MCP tool. Every @click.option() becomes a typed input parameter. No boilerplate, no schema writing, no maintenance burden when your CLI changes.

Transport options:

• serve — stdio (for Claude Desktop, local IDEs)
• serve-http — HTTP+SSE (for remote access, web integrations)
• serve-http-streamable — Streamable HTTP (modern stateless transport)

Distribution tip: Submit to Glama MCP Directory, mcp.so, and the Awesome MCP list. These are where MCP users discover new servers. click-to-mcp is already listed in all three.

Channel 2: Claude Code Plugin

New in click-to-mcp v0.5.0

Claude Code is Anthropic's CLI for coding agents. It has its own plugin system with a directory of 1,715+ plugins. Users install plugins with a single slash command — no config editing, no JSON schemas to write.

click-to-mcp now ships as a Claude Code plugin out of the box:

/plugin install click-to-mcp@Coding-Dev-Tools/click-to-mcp

That's it. The plugin runs click-to-mcp's built-in demo CLI as an MCP server, exposing greet, calculate, and config tools. Users can then configure it to wrap their own CLIs.

How to Package Your Own CLI as a Plugin

The plugin structure is simple — two files in your repo:

.claude-plugin/
  plugin.json       # Metadata: name, description, version, keywords
.mcp.json           # MCP server config: command, args, env

The plugin.json for click-to-mcp looks like this:

{
  "name": "click-to-mcp",
  "description": "Auto-wrap any Click/typer CLI as an MCP server",
  "version": "0.5.0",
  "author": { "name": "DevForge" },
  "keywords": ["mcp", "cli", "click", "typer", "devops"]
}

And the .mcp.json tells Claude Code how to start the server:

{
  "mcpServers": {
    "click-to-mcp": {
      "command": "uvx",
      "args": ["click-to-mcp", "serve", "click_to_mcp.demo:cli"]
    }
  }
}

Why this matters: The Claude Code plugin directory is the first place AI-focused developers look for tools. If your CLI isn't there, it doesn't exist for that audience. The install flow is friction-free compared to manual MCP config — one command vs. editing JSON files by hand.

Channel 3: Agent Skill (The Broadest Reach)

New in click-to-mcp v0.5.0

The Agent Skills specification (19,389 stars) is the emerging open standard for extending AI agents. Originally developed by Anthropic, now adopted across Claude Code, Cursor, Copilot, and independent agents like Hermes.

A Skill = a folder with SKILL.md + optional scripts/references/assets. Progressive disclosure: discovery → activation → execution. Cross-product reuse: build once, run on any Skills-compatible agent.

click-to-mcp now ships as an Agent Skill:

skill/
  click-to-mcp/
    SKILL.md          # Instructions the agent reads
    references/
      discovery-guide.md
      transports.md

When an agent loads the skill, it gets full instructions on how to use click-to-mcp — when to use it, how to install it, which transport to pick, and common troubleshooting steps. The agent reads SKILL.md and knows exactly what to do without the user explaining anything.

How to Create an Agent Skill for Your CLI

1. Create a skill/your-tool-name/ folder in your repo.

2. Write SKILL.md with YAML frontmatter + markdown body:

---
name: your-tool-name
description: One-line description. When the agent should use this skill.
---

# your-tool-name: What It Does

## Overview
Explain what the tool does and why an agent would use it.

## Quick Start
Step-by-step install and usage instructions.

## Common Patterns
Real examples the agent can follow.

## Troubleshooting
Known issues and fixes.

3. Add reference files for deep knowledge the agent might need.

4. Submit to agentskills.io for discovery.

Why this matters: Agent Skills work across every major agent platform, not just Claude Code. A single SKILL.md makes your CLI discoverable and usable by any agent that speaks the Skills protocol. This is the broadest distribution channel available today.

Which Channel Should You Prioritize?

Start with MCP (it's the baseline), then add the other two in order of your audience:

The good news: with click-to-mcp, the MCP server is automatic. The plugin and skill are just two files each. Total effort to cover all three channels: under an hour.

The Bigger Picture: CLIs Are the New API Surface

Six months ago, making your CLI "AI-ready" meant writing a custom MCP server from scratch — hundreds of lines of boilerplate that broke every time you added a CLI option. Now the bar is one command: click-to-mcp serve.

But being AI-ready isn't enough anymore. You also have to be AI-discoverable. That means showing up in the plugin directories and skill registries where AI agent users search for tools. The three-channel strategy isn't optional — it's table stakes for any CLI that wants adoption in 2026.

If you're building Click or Typer CLIs, click-to-mcp handles all three channels. Star the repo — it helps other devs find it.

Get DevForge Updates

New tools, deep dives, and AI agent dev stories — delivered to your inbox.

Star us on GitHub · Introducing click-to-mcp → · Usage Guide → · View Pricing → · Documentation →