Introduction to Model Context Protocol · lesson 4 of 6

MCP Prompts

Parameterized prompt templates exposed by an MCP server. User-controlled — typically surfaced as slash commands in the host UI.

Concept

A prompt is a named, parameterized message template. The server returns a ready-to-send list of messages when the user invokes it.

Example: a /review-pr prompt in a GitHub MCP server takes a pr_url argument and returns a structured review prompt populated with the PR body and diff.

Definition

from mcp.types import Prompt, PromptArgument

@server.list_prompts()
async def list_prompts():
    return [
        Prompt(
            name="review_pr",
            description="Generate a code-review prompt for a GitHub pull request",
            arguments=[
                PromptArgument(name="pr_url", description="GitHub PR URL", required=True),
                PromptArgument(name="focus", description="Optional focus area", required=False),
            ],
        )
    ]

@server.get_prompt()
async def get_prompt(name: str, arguments: dict):
    if name == "review_pr":
        pr = await github.get_pr(arguments["pr_url"])
        return {
            "messages": [
                {"role": "user", "content": [{"type": "text", "text": f"Review this PR:\n{pr.body}\n\nDiff:\n{pr.diff}"}]}
            ]
        }

Argument autocomplete

A server can implement completion/complete so the host UI offers autocomplete as the user types arguments. Example: typing /review_pr pr_url= could autocomplete from a list of recent open PRs.

Prompts vs tools

Aspect	Prompt	Tool
Invoked by	User (slash command)	Model
Returns	Messages to seed the conversation	Tool result
Side effects	No	Often
Discovery	Slash command menu	LLM reads tool list

CCA-F angle

The exam tests whether you can pick the right primitive: