Resources

Resources expose content to LLMs, enabling them to read from:

• Local documents and files
• Databases
• APIs
• Knowledge bases

Tools

Tools let the LLM take action via the MCP server:

• Write to disk
• Send emails
• Run scripts or database queries
• Trigger remote actions

Prompts

Reusable prompt templates ensure structured and consistent LLM behavior. They can:

• Guide specialized agents
• Chain workflows
• Enable multi-step reasoning

Let’s look at how simple it is to expose tools and resources to LLMs using FastMCP, a Python SDK for building MCP servers.

First, initialize the server:

from mcp.server.fastmcp import FastMCP

from mcp.server.fastmcp import FastMCP

Create Notes

@mcp.tool(description=”Create a new note. Only save. Don’t modify existing notes. Supports emojis.”)
def add_note(message: str) -> str:
# Save message to next available file

Read All Notes

@mcp.tool(description=”Return all notes. Only read them. Do not summarize or edit.”)
def read_notes() -> str:
# Read and return all note contents

Edit a Note 

@mcp.tool(description=”Edit an existing note by its number. Only do this when explicitly asked to edit.”)
def edit_note(index: int, new_message: str) -> str:
# Update a specific note by number

Access Notes as Resources

@mcp.resource(“notes://{index}”, description=”Read a specific note by number. Do not modify it.”)
def get_note_by_index(index: int) -> str:
# Return contents of note_{index}.txt

Summarize Notes

@mcp.prompt(description=”Generate a summary prompt of all notes, without changing them.”)
def note_summary_prompt() -> str:
# Format and return a summarization prompt

This compact example shows how you can turn a folder of text files into an AI-powered sticky note assistant with structured context and callable actions.