mcp-learning

A structured Python learning journey for a senior TypeScript/Angular engineer building MCP agents with the Anthropic SDK.

Goal

Learn enough Python to confidently build production-quality MCP (Model Context Protocol) servers and Anthropic SDK integrations — not Python in general, but Python for this specific use case.

Phase Breakdown

phase1-basics/ — Python Syntax & Types

Core language mechanics for someone who already knows TypeScript well. Focus on what's different, not what's familiar.

• Type hints vs. TypeScript generics
• Dataclasses vs. interfaces
• List/dict comprehensions
• async/await (same concept, different runtime model)
• Exceptions and context managers

phase2-apis/ — HTTP & FastAPI

Building and consuming HTTP APIs the Python way.

• httpx for async HTTP requests (the fetch/axios equivalent)
• FastAPI for building typed REST endpoints
• Pydantic models for runtime validation (think Zod + types in one)

phase3-llm/ — Anthropic SDK & Tool Use

Direct SDK usage before abstracting to MCP.

• anthropic SDK basics: messages, streaming
• Tool use / function calling
• Prompt caching
• Multi-turn conversations

phase4-mcp/ — MCP Server Builds

Putting it all together.

• MCP server scaffolding with mcp SDK
• Exposing tools, resources, and prompts
• Connecting to Claude Desktop / Claude Code
• Testing MCP servers locally

Setup

# Install dependencies (uses uv)
uv sync

# Copy and fill in your API key
cp .env.example .env

# Run the phase 1 starter
uv run phase1-basics/hello.py

Prerequisites

• Python 3.11+
• uv for package management
• Anthropic API key (needed from phase 3 onward)