Use a Server-Provided Prompt

A prompt is neither a tool nor a resource

So far, you have used two kinds of capability: a tool, which performed a search, and a resource, which returned identified information.

The Practice Resource Desk exposes a third kind: plan_a_research_session.

This is an MCP prompt. It provides a reusable way to structure a request.

It does not search the fictional collection. It does not return the overview resource. And it does not, by itself, ask an AI model to answer anything.

Three words to learn

Step 1 — See what the server offers

Start the Inspector and connect to practice-resource-desk as before, then click the Prompts tab.

One prompt is listed. Click Plan a research session, then stop — do not retrieve it yet.

The technical name is not on screen here. On the Tools tab, the Inspector showed both Search resources and search_resources. On the Prompts tab it shows only the readable title, so you will not see plan_a_research_session anywhere in this view. It is still the name the server uses.

The prompt has a blank to fill in

plan_a_research_session is reusable because part of it is left open. The prompt expects a topic.

The topic is an argument: information supplied to fill in part of the template.

The red asterisk marks it as required, and the Inspector enforces that before anything is sent — Get Prompt stays grayed out and unclickable until you type something into the field. This is the same pattern you saw with Execute Tool in Module 2.

Step 2 — Supply the topic

Type source evaluation into the field, which shows the placeholder Enter topic…

Stop before clicking anything. Ask yourself two questions:

• What part came from the server? The reusable structure — the instructions, the numbered steps, the restriction to the desk’s own resources.
• What part did you supply? Only source evaluation.

Notice how little you contributed, and how much the server decided.

Step 3 — Ask the server for the filled-in prompt

Click Get Prompt, which is now enabled.

The form is replaced by a panel headed Messages:

The returned prompt begins:

I want to plan a short research session on: source evaluation

Use only the fictional resources available from the Practice Resource Desk.

Please:
1. Search the desk for resources related to this topic.
2. Tell me which ones look relevant and why.
3. Suggest an order to use them in.
4. Say plainly what the desk does not cover, rather than filling the gap.

Above that text sits a short line reading [0] role: user. Inspector shows how this content would be presented as part of a prompt. You do not need to memorize the technical structure yet.

To go back to the form and try a different topic, click the × beside Messages.

What happened?

1. You selected a prompt the server exposes.
2. You supplied the topic source evaluation.
3. Inspector requested the prompt from the Practice Resource Desk.
4. The server filled the topic into its reusable template.
5. Inspector displayed the resulting prompt content.

Tool, resource, or prompt?

All three can return text. What matters is the kind of capability you asked the server to provide.

What to notice

The server supplied structure; you supplied context.

plan_a_research_session was reusable because the topic could change while the overall prompt structure stayed the same.

You retrieved a prompt, but you did not run the prompt through an AI model. That distinction becomes important when MCP is connected to an AI application later in the tutorial.

It is worth reading what came back on its own terms. The text you retrieved contains instructions — it tells a model what to do, what to use, and what not to invent. Before using a server-provided prompt with an AI model, it is worth reading what the server actually returned.

Checkpoint

You are ready for Module 5 if you can:

• explain an MCP prompt without calling it a tool;
• identify the reusable part and the supplied argument;
• retrieve a prompt through Inspector;
• inspect the filled-in prompt content;
• distinguish prompt retrieval from AI generation; and
• compare tools, resources, and prompts in ordinary language.