Homework

Task 1 — Run the reference mcp-server-everything

The everything server is the canonical “kitchen-sink” reference: it exposes a deliberately varied set of tools, prompts, and resources designed to exercise every corner of the protocol. Running it under Inspector is the fastest way to feel what a fully-populated MCP server looks like — useful context before you start growing your own from zero.

Launch it under mcp dev via npx:

npx -y @modelcontextprotocol/server-everything

Then point Inspector at that stdio process (the easiest way is to run npx @modelcontextprotocol/inspector npx -y @modelcontextprotocol/server-everything so Inspector spawns the server itself). In the Inspector UI, open the Tools tab and read through the list. Pick one that looks interesting — for example a math tool or one that returns a formatted resource — and call it with sensible arguments. Note the JSON the server returns.

Task 2 — Side-by-side rewrite: low-level ↔ FastMCP

Slide 3 of the Lesson 03 deck shows the hello tool implemented in the low-level Server style — explicit list_tools /call_tool handlers, manual JSON schema, manual TextContent wrapping. Your job: rewrite the same tool in FastMCP (the decorator-driven high-level wrapper) — or, if you prefer, do it the other direction (start from the FastMCP version and write the low-level equivalent). Either direction is fine; the point is to feel the difference.

Drop both versions into your writeup side by side — literally two code blocks next to each other, or a two-column table. Then run each one separately through mcp dev and confirm both show a tool named hello with the same input shape and output behaviour.

Skeleton to start from (FastMCP side):

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("hello-demo")

@mcp.tool()
def hello(name: str) -> str:
    """Say hello to NAME."""
    return f"Hello, {name}!"

if __name__ == "__main__":
    mcp.run()

Task 3 — Read one reference server end-to-end

Pick one of the official Python reference servers from modelcontextprotocol/servers — either fetch (HTTP requests) or sqlite (local database queries). Find its server.py (or whatever the equivalent entry point is — check pyproject.toml for the [project.scripts] table if you're not sure) and read it top to bottom.

In exactly ~100 words, write a short writeup covering:

• What tools the server exposes (name them).
• What each tool does, in one sentence.
• One design choice in the source that surprised you — an idiom, an abstraction, a defensive check, anything you wouldn't have written yourself the first time.