Skip to content

05 · FAQ

The full version of the short FAQ in the README. Answers assume you've done the steps — click the link in each answer to go deeper.

General

What exactly is MCP? An open standard that lets AI apps (Claude, ChatGPT, VS Code, Cursor) connect to external tools and data. "USB-C for AI": build the integration once, plug it into any MCP-capable app. → 01-concepts.md

Is MCP a programming language or a framework? Neither. It's a protocol — a defined format for messages (JSON-RPC 2.0) and how they travel (transport). The Python SDK turns it into decorators you write normal functions behind.

Do I need an AI model to follow this tutorial? No, not for Steps 1–6. You write servers and a tiny client, and you watch real JSON-RPC traffic — no LLM involved. Only Step 7 (the host) needs the free Groq API key.

Who owns MCP? It started at Anthropic and is now developed in the open under the Model Context Protocol organization, with contributions from Anthropic, OpenAI, Google, Microsoft, and many others.

Architecture

What's the difference between Host, Client, and Server? - Host = the AI app you use (Claude Desktop, VS Code, your Step 7 app). - Client = a component the host creates for each server; maintains the connection. - Server = your program that exposes tools, resources, and prompts.

One host can talk to many servers at once — each via its own client. → 01-concepts.md

What's the difference between a tool, a resource, and a prompt? - Tool — something the model can call to act (calculate, fetch, send). - Resource — read-only data the app loads as context. - Prompt — a reusable instruction template the user invokes.

Rule of thumb: does something → tool; is data → resource; is a recipe → prompt.

Which transport should I use? - stdio — server lives on the same machine; the client spawns it and talks over stdin/stdout. Used throughout this tutorial. - Streamable HTTP — server is remote; HTTP POST + Server-Sent Events, with auth. Same JSON-RPC messages, different plumbing.

Why can't I print() in a stdio server? Because stdout is the protocol channel. A stray print() corrupts the JSON-RPC stream and the server silently breaks. Use logging (stderr) instead. This is the golden rule — see 01-concepts.md.

Function calling & MCP

Isn't MCP just function calling? Function calling = the grammar an LLM uses to request a tool. MCP = the transport + discovery standard that carries tools and data between apps and servers. They're complementary. → 04-mcp-vs-alternatives.md

Why does Step 7 (the host) use Groq function calling and MCP? The Groq API decides which tool to call (function calling). The MCP host (mcp_host.py) actually runs it on the server and returns the result. Two different layers working together — exactly how real AI apps work.

Can I use the host UI with a different LLM provider? Yes, with a small change: swap the groq client in chat_engine.py for any OpenAI-compatible client (they share the function-calling message format). Anthropic and Gemini use a similar but slightly different format.

Using your servers

Why does python server_basic.py just sit there? It's working correctly. A stdio server doesn't run on its own — a client must spawn it and start the conversation (Step 2's script or the Inspector in Step 3).

Why does the Inspector look different from YouTube tutorials? This project runs Inspector v2 (tabs), while older tutorials show v1 (one screen). They inspect the same protocol. → 03-inspector-v2-guide.md

Do the config files contain my paths? No — the repo is path-independent. inspector.json and .vscode/mcp.json use relative paths so they work for anyone who opens the folder. claude_desktop_config.json uses a <YOUR-PATH> placeholder because Claude Desktop requires absolute paths — replace it before copying.

The weather tool says "Unable to fetch". What now? Usually no internet access, an Open-Meteo rate limit (wait a minute), or a temporarily blocked User-Agent. The server logs the underlying error to stderr.

Troubleshooting recap

Full table in the README. The four most common gotchas:

Symptom Likely cause
No module named 'mcp' venv not activated
Server starts then instantly exits print() in server code (stdout pollution)
Client hangs Config path wrong / server not spawnable
Groq call failed: ... 401 Missing or wrong Groq API key

Learn more