MCP Zero to Hero

An open-source monorepo for learning, building, and scaling Model Context Protocol (MCP) servers — from a Hello World tool to research-grade agentic systems.

What Is MCP?

The Model Context Protocol is an open standard that lets AI host applications (like Claude Desktop or IDE copilots) discover and use external capabilities through a single, typed, transport-agnostic protocol.

flowchart LR
    subgraph Hosts["Host Applications"]
        H1[Claude Desktop]
        H2[IDE Copilot]
        H3[Custom App]
    end
    subgraph Protocol["MCP Protocol\n(JSON-RPC 2.0)"]
    end
    subgraph Servers["Your MCP Servers"]
        S1[Files]
        S2[Databases]
        S3[APIs]
        S4[Custom Logic]
    end

    H1 <-->|stdio or SSE| Protocol
    H2 <-->|stdio or SSE| Protocol
    H3 <-->|stdio or SSE| Protocol
    Protocol <--> S1
    Protocol <--> S2
    Protocol <--> S3
    Protocol <--> S4

    style Hosts fill:#dbeafe
    style Protocol fill:#d1fae5
    style Servers fill:#fef3c7

MCP servers expose three capability families:

| Family | What it does | Example | | ------------- | ------------------------------ | -------------------------------- | | Tools | Executable operations | create_issue, query_table | | Resources | URI-addressed read-only data | schema://main/users | | Prompts | Reusable interaction templates | code_review, persona_greeter |

Repository Layout

MCP-Zero-To-Hero/
├── theory/          ← 7 conceptual chapters: architecture → security
├── modules/         ← shared helpers, templates, config snippets
└── projects/
    ├── beginner/    ← Projects 01-05: fundamentals
    ├── intermediate/← Projects 06-12: real integrations
    ├── advanced/    ← Projects 13-17: compound workflows
    └── research/    ← Projects 18-20: agentic / sampling

Learning Path

flowchart LR
    T[theory/\n7 chapters] --> B[Beginner\n01-05]
    B --> I[Intermediate\n06-12]
    I --> A[Advanced\n13-17]
    A --> R[Research\n18-20]

    style T fill:#e0e7ff
    style B fill:#dbeafe
    style I fill:#d1fae5
    style A fill:#fef3c7
    style R fill:#fee2e2

| Track | Projects | Core skills | | ---------------- | -------- | ----------------------------------------------------- | | Beginner | 01–05 | Tools, resources, prompts, input validation | | Intermediate | 06–12 | Real APIs, filesystems, databases, automation | | Advanced | 13–17 | Vector DBs, cloud infra, pipelines, transport choices | | Research | 18–20 | Host sampling, self-healing loops, semantic search |

Quick Start

Prerequisites

• Node.js 20+ and npm 10+ (for TypeScript projects)
• Python 3.10+ and pip (for Python projects)
• A host app such as Claude Desktop

Run Your First Server (TypeScript)

cd projects/beginner/01-hello-world-server-ts
npm install
npm run build
npm start

Run Your First Server (Python)

cd projects/beginner/02-local-file-system-reader-py
pip install -e .
python3 server.py

Register With Claude Desktop

Each project README includes the exact config snippet. Paste it into claude_desktop_config.json:

{
  "mcpServers": {
    "hello-world-server": {
      "command": "node",
      "args": ["/absolute/path/to/dist/index.js"]
    }
  }
}

Repository Rules

• TypeScript projects use @modelcontextprotocol/sdk (official SDK).
• Python projects use the official mcp package.
• stdio is the default transport for all projects.
• Every project includes the exact claude_desktop_config.json snippet.
• stdout is reserved for MCP protocol traffic only — all logs go to stderr.

Contents

• theory/README.md — conceptual foundation, 7 chapters
• projects/README.md — full 20-project catalog with descriptions
• modules/README.md — shared utilities and documentation templates

Intended Audience

• Beginners learning MCP concepts for the first time
• Engineers building local MCP servers for personal or team use
• DevRel teams teaching MCP with production-ready examples
• Advanced practitioners exploring agentic and research-oriented MCP workflows