Homework — Lesson 01

Task 1 — Explore another official server

Pick one of the following official MCP servers and add it to your Claude Desktop config alongside the filesystem server you set up in the lab:

• mcp-server-fetch — fetches a URL and returns its contents as text. Good for "summarise this page" prompts.
• mcp-server-sqlite — opens a SQLite database file and exposes read/query tools. Good for "what's in this DB?" prompts.
• mcp-server-everything — a kitchen-sink reference server that exercises every part of the protocol (tools, resources, prompts, sampling). Good if you want to see the full surface area.

Steps:

1. Add a second entry under mcpServers in claude_desktop_config.json (do not remove the filesystem server — both should be live at the same time).
2. Fully quit and restart Claude Desktop.
3. Ask Claude one question that requires the new server (for example, with fetch: "Fetch https://modelcontextprotocol.io and summarise the homepage in three bullets").
4. Watch the tool-call block in the chat — confirm the right server handled it.

✓ Self-check: a screenshot of Claude Desktop's UI showing the tool call and its result for the new server.

Task 2 — Write a peer-friendly MCP intro

Imagine a developer friend who has used the Claude API for a year and writes their own tool-use scaffolding by hand, but has never heard of MCP. Write a ~200-word introduction for them — clear, concrete, no jargon dumps.

Cover, at minimum:

• The problem MCP solves (be specific — what pain are they feeling that they don't yet know has a name?).
• The three primitives (tools, resources, prompts) — one sentence each is fine.
• One realistic use case where MCP makes their life easier than rolling their own tool-use shim.

Keep it conversational. You're not writing marketing copy; you're explaining a useful pattern to a peer.

✓ Self-check: your write-up includes the words client, server, protocol, and at least one of tools, resources, or prompts.

Task 3 — Map the M×N problem to your job

Pick two LLM clients you already use or could realistically adopt (Claude Desktop, Cursor, a custom agent, ChatGPT desktop, Zed, Continue, whatever) and three tools or capabilities you'd want them to have access to (your team's wiki, a Postgres database, an internal API, your calendar, a code search index, etc.).

1. Sketch the integration matrix without MCP: how many separate shims, plugins, or adapters would you need? Write them out as bullet points of the form "I would need a Y-specific adapter for X". (With 2 clients × 3 tools you'll have at least 6 lines.)
2. Now describe how MCP collapses that matrix: which side owns which piece, and what each integration becomes.
3. Be honest about the edges — auth, deployment, who runs the server — where the simplification has limits.

✓ Self-check: you can replace ≥5 lines of "I would write a Y-specific shim for X" with one MCP server that both clients use.