← Back to Blog
tutorialecosystemoverview

The Complete MCPM Ecosystem Guide

How mcpm fits into the broader MCP landscape — clients, servers, registries, and the tools that make it all work together.

·4 min read·xapable

The Complete MCPM Ecosystem Guide

mcpm doesn't exist in isolation. This guide maps the entire MCP landscape and shows how everything fits together.

The Big Picture

┌─────────────────────────────────────────────────┐
│                  MCP Ecosystem                    │
│                                                   │
│  ┌──────────┐   ┌──────────┐   ┌──────────┐     │
│  │  Claude  │   │  Cursor  │   │ Windsurf │     │
│  │ Desktop  │   │   IDE    │   │          │     │
│  └────┬─────┘   └────┬─────┘   └────┬─────┘     │
│       │               │               │           │
│       └───────────────┬───────────────┘           │
│                       │                            │
│                  MCP Protocol                      │
│                       │                            │
│       ┌───────────────┼───────────────┐           │
│       │               │               │           │
│  ┌────┴─────┐   ┌─────┴────┐   ┌─────┴────┐     │
│  │  mcpm    │   │ mcpm.sh  │   │  Local   │     │
│  │ (registry)│   │(launcher)│   │ Servers  │     │
│  └──────────┘   └──────────┘   └──────────┘     │
└─────────────────────────────────────────────────┘

Layer 1: MCP Clients

These are the AI applications that consume MCP tools:

Client Type Best For
Claude Desktop Desktop app General-purpose AI assistant
Cursor IDE AI-powered coding
Windsurf IDE Agentic coding workflows
Continue IDE extension Open-source AI coding
Claude Code CLI Terminal-based AI coding

All of them speak the MCP protocol. Any tool you publish to mcpm works with all of them.

Layer 2: MCP Protocol

The Model Context Protocol is the standard that connects clients to servers:

  • Open standard — anyone can implement it
  • JSON-RPC over stdio — simple, universal
  • Tool discovery — clients ask servers what they can do
  • Streaming — coming in MCP 2.0

Layer 3: mcpm (The Registry)

This is mcpm.dev — the central registry for MCP tools. Like npm for AI agents.

What mcpm does:

  • Hosts tool listings: name, description, README, versions
  • Search: semantic search across all published tools
  • Publishing: one-command publish from the CLI
  • Discovery: browse, filter, sort by popularity

What mcpm does NOT do:

  • Does not run tools — that's the client's job
  • Does not manage local installations — that's mcpm.sh's job
  • Does not host code — tools live on npm/GitHub

Layer 4: Tool Runners & Launchers

mcpm.sh

The local MCP server manager. Handles:

  • Installing and running MCP servers locally
  • Managing server configurations
  • Process lifecycle (start, stop, restart)

Docker-based runners

Some tools run in containers for isolation:

<span class="hljs-punctuation">{</span>
  <span class="hljs-attr">&quot;mcpServers&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
    <span class="hljs-attr">&quot;my-tool&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
      <span class="hljs-attr">&quot;command&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;docker&quot;</span><span class="hljs-punctuation">,</span>
      <span class="hljs-attr">&quot;args&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">[</span><span class="hljs-string">&quot;run&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-string">&quot;-i&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-string">&quot;--rm&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-string">&quot;my-tool-image&quot;</span><span class="hljs-punctuation">]</span>
    <span class="hljs-punctuation">}</span>
  <span class="hljs-punctuation">}</span>
<span class="hljs-punctuation">}</span>

npx-based (Most Common)

Tools published to npm run directly:

npx -y weather-mcp

How Discovery Works

  1. You search on mcpm: mcpm-dev search weather
  2. mcpm returns matching tools with descriptions
  3. You read details: mcpm-dev info weather-mcp
  4. You install: mcpm-dev add weather-mcp (delegates to npm)
  5. You configure your MCP client to run the tool
  6. The client launches the tool as a child process
  7. The client discovers available tools via tools/list
  8. Your AI agent can now call those tools

How Publishing Works

  1. You build an MCP server (Node.js, Python, etc.)
  2. You publish to npm: npm publish
  3. You publish to mcpm: mcpm-dev publish
  4. mcpm reads your package.json and README.md
  5. Your tool is now discoverable on mcpm.dev
  6. Users find it via search, install, and use it

The Ecosystem Philosophy

mcpm is one piece of a larger puzzle:

mcpm.dev  →  discovery + registry
mcpm.sh   →  local management
npm       →  package hosting
Clients   →  AI-powered usage

Each piece does one thing well. Together, they make MCP tools accessible to everyone.

Getting Started

New to MCP? Here's your path:

  1. Install a client — Claude Desktop (easiest start)
  2. Browse mcpm.dev — see what tools exist
  3. Install a tool — mcpm-dev add weather-mcp
  4. Use it — ask your AI agent to use the tool
  5. Build your own — follow our "Build Your First MCP Server" tutorial
  6. Publish — share it with the world on mcpm

The MCP ecosystem is growing fast. mcpm is your entry point — the npm for the AI age.

#MCP #Tutorial #Ecosystem #mcpm #AIAgents

← Back to all posts