Prompts en servidores MCP

Introducción a Model Context Protocol (MCP)

James Chapman

AI Curriculum Manager, DataCamp

¿Qué son los prompts de MCP?

Prompts: plantillas reutilizables que optimizan los LLM para tareas concretas

Flujos y comportamientos
Reduce la carga de prompt para el usuario

La necesidad de hacer prompting

El prompt del conversor de zonas horarias

You are a timezone conversion engine.

Your task is to:
1. Extract the source datetime from the user's natural language input.
2. Identify the source timezone (explicit or inferred).
3. Convert the datetime into the target timezone.

Rules:
- If the date is ambiguous (e.g., "next Friday"), resolve it relative to the provided datetime.
- If the input cannot be resolved confidently, seek clarification.

→ Expón este prompt para la tarea de conversión de zonas horarias

Definir prompts del servidor MCP

@mcp.prompt(title="Timezone Conversion")

def convert_timezone_prompt(user_input: str) -> str:

    return f"""You are a timezone conversion engine.

Your task is to:
1. Extract the source datetime...

Rules:
- If the date is ambiguous (e.g., "next Friday"), resolve it relative to the provided datetime.
- If the input cannot be resolved confidently, seek clarification.

    User's timezone conversion request: {user_input}"""

Servidor MCP local: timezone_server.py

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("Timezone Converter")

# Tools and resources from before...

@mcp.prompt(title="Timezone Conversion")
def convert_timezone_prompt(timezone_request: str) -> str:
    # ...

if __name__ == "__main__":
    mcp.run(transport="stdio")

Cliente: listar prompts

async def list_prompts():
    """List all available prompts from the MCP server."""
    params = StdioServerParameters(command=sys.executable, args=["timezone_server.py"])

    async with stdio_client(params) as (reader, writer):
        async with ClientSession(reader, writer) as session:
            await session.initialize()

            # List available prompts
            prompts = await session.list_prompts()
            print(f"Available prompts: {[p.name for p in prompts.prompts]}")

Cliente: listar prompts

print(asyncio.run(list_prompts()))

Available prompts: ['convert_timezone_prompt']

El name del prompt no es lo mismo que el title

@mcp.prompt(title="Timezone Conversion")
def convert_timezone_prompt(timezone_request: str) -> str:
    # ...

Cliente: listar prompts

print(asyncio.run(list_prompts()))

Available prompts: ['convert_timezone_prompt']

El name del prompt no es lo mismo que el title

# List available prompts
prompts = await session.list_prompts()
print(f"Available prompts: {[p.name for p in prompts.prompts]}")

Se usa el atributo .name en el cliente (no .title)

Cliente: recuperar prompts

async def read_prompt(user_input: str, prompt_name: str = "convert_timezone_prompt") -> str:
    """Retrieve a prompt from the MCP server with user input."""
    params = StdioServerParameters(command=sys.executable, args=["timezone_server.py"])

    async with stdio_client(params) as (reader, writer):
        async with ClientSession(reader, writer) as session:
            await session.initialize()

            prompts = await session.list_prompts()

            # Retrieve the prompt
            if prompts.prompts:
                prompt = await session.get_prompt(prompt_name, arguments={"timezone_request": user_input})

                print(f"Prompt result: {prompt.messages[0].content.text}")

            return prompt.messages[0].content.text

print(asyncio.run(read_prompt(user_input="It is 9:50 AM in the UK in January. What time is
    it in Lisbon, Portugal?")))

You are a timezone conversion engine.

Your task is to:
1. Extract the source datetime from the user's natural language input.
2. Identify the source timezone (explicit or inferred).
3. Convert the datetime into the target timezone.

Rules:
- If the date is ambiguous (e.g., "next Friday"), resolve it relative to the...
- If the input cannot be resolved confidently, seek clarification.

User's timezone conversion request: It is 9:50 AM in the UK in January. What time is it in
Lisbon, Portugal?

¡Vamos a practicar!

Introducción a Model Context Protocol (MCP)