Cómo crear un agente de IA con DSPy y el servidor MCP de ClickHouse

En esta guía, aprenderás a crear un agente de IA con DSPy que puede interactuar con el Playground de SQL de ClickHouse mediante el servidor MCP de ClickHouse.

Requisitos previos

Debes tener Python instalado en el sistema.
Debes tener pip instalado en el sistema.
Necesitarás una clave de API de Anthropic o una clave de API de otro proveedor de LLM.

Puedes ejecutar los siguientes pasos desde el REPL de Python o mediante un script.

Notebook de ejemploEste ejemplo está disponible como notebook en el repositorio de ejemplos.

Instale las bibliotecas

Ejecute los siguientes comandos con pip para instalar las bibliotecas necesarias:

pip install -q --upgrade pip
pip install -q dspy
pip install -q mcp

Configura las credenciales

A continuación, deberás proporcionar tu clave de API de Anthropic:

import os
os.environ["ANTHROPIC_API_KEY"] = getpass.getpass("Enter Anthropic API Key:")

Usar otro proveedor de LLMSi no tienes una clave de API de Anthropic y quieres usar otro proveedor de LLM, puedes consultar las instrucciones para configurar tus credenciales en la documentación de DSPy

A continuación, define las credenciales necesarias para conectarte al Playground de SQL de ClickHouse:

env = {
    "CLICKHOUSE_HOST": "sql-clickhouse.clickhouse.com",
    "CLICKHOUSE_PORT": "8443",
    "CLICKHOUSE_USER": "demo",
    "CLICKHOUSE_PASSWORD": "",
    "CLICKHOUSE_SECURE": "true"
}

Inicializar el servidor MCP

Ahora configura el servidor MCP de ClickHouse para que apunte al Playground de SQL de ClickHouse.

from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client
import dspy

server_params = StdioServerParameters(
    command="uv",
    args=[
        'run',
        '--with', 'mcp-clickhouse',
        '--python', '3.13',
        'mcp-clickhouse'
    ],
    env=env
)

Inicializar el LLM

A continuación, inicialice el LLM con la siguiente línea:

dspy.configure(lm=dspy.LM("anthropic/claude-sonnet-4-20250514"))

Ejecutar el agente

Por último, inicialice y ejecute el agente:

class DataAnalyst(dspy.Signature):
    """You are a data analyst. You'll be asked questions and you need to try to answer them using the tools you have access to. """

    user_request: str = dspy.InputField()
    process_result: str = dspy.OutputField(
        desc=(
            "Answer to the query"
        )
    )

from utils import print_dspy_result

async with stdio_client(server_params) as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()
        tools = await session.list_tools()

        dspy_tools = []
        for tool in tools.tools:
            dspy_tools.append(dspy.Tool.from_mcp_tool(session, tool))

        react = dspy.ReAct(DataAnalyst, tools=dspy_tools)
        result = await react.acall(user_request="What's the most popular Amazon product category")
        print_dspy_result(result)

Response

================================================================================
🤖 DSPy ReAct Result
================================================================================

📍 STEP 1
----------------------------------------
🧠 THINKING: I need to find information about Amazon product categories and determine which one is most popular. First, I should explore what databases are available to see if there's any Amazon-related data.

🔧 TOOL: list_databases

📊 RESULT:
   amazon
bluesky
country
covid
default
dns
environmental
food
forex
geo
git
github
hackernews
imdb
logs
metrica
mgbench
mta
noaa
nyc_taxi
nypd
ontime
opensky
otel
otel_v2
pypi
random
reddit
rubygems
sta...

📍 STEP 2
----------------------------------------
🧠 THINKING: Great! I can see there's an "amazon" database available. This is exactly what I need to find information about Amazon product categories. Let me explore the tables in the amazon database to see what data is available.

🔧 TOOL: list_tables
   Args: {'database': 'amazon'}

📊 RESULT:
   {
  "database": "amazon",
  "name": "amazon_reviews",
  "comment": "",
  "columns": [
    {
      "name": "review_date",
      "type": "Date",
      "default_type": "",
      "default_expression": "",...

📍 STEP 3
----------------------------------------
🧠 THINKING: Perfect! I found the amazon_reviews table which contains a product_category column. This table has over 150 million rows of Amazon review data, which should give me a good representation of product popularity. To find the most popular Amazon product category, I need to count the number of reviews per category and order them by count in descending order. The category with the most reviews would be considered the most popular.

🔧 TOOL: run_select_query
   Args: {'query': 'SELECT product_category, COUNT(*) as review_count FROM amazon.amazon_reviews GROUP BY product_category ORDER BY review_count DESC LIMIT 10'}

📊 RESULT:
   1. Books: 19,530,930 reviews
   2. Digital_Ebook_Purchase: 17,622,107 reviews
   3. Wireless: 9,001,567 reviews
   4. PC: 6,908,245 reviews
   5. Home: 6,221,321 reviews
   ... and 5 more

📍 STEP 4
----------------------------------------
🧠 THINKING: Perfect! I have found the answer to the user's question. Based on the Amazon reviews data, I can clearly see that "Books" is the most popular Amazon product category with 19,530,930 reviews, followed by "Digital_Ebook_Purchase" with 17,622,107 reviews. The data shows a clear ranking of product categories by popularity based on the number of reviews each category has received. I now have all the information needed to answer the user's question.

🔧 TOOL: finish

📊 RESULT:
   Completed.

🎯 REASONING
----------------------------------------
To answer the question about the most popular Amazon product category, I accessed the Amazon database and found a table called "amazon_reviews" with over 150 million review records. This table contains a "product_category" column that allows me to analyze product popularity based on review volume. I ran a query to count the number of reviews per product category and ordered them by count in descending order. The assumption here is that categories with more reviews indicate higher popularity, as more reviews typically correlate with more purchases and customer engagement.

✅ FINAL RESULT
----------------------------------------
Based on Amazon review data containing over 150 million records, the most popular Amazon product category is **Books** with 19,530,930 reviews. 

Here are the top 10 most popular Amazon product categories by review count:

1. **Books** - 19,530,930 reviews
2. **Digital_Ebook_Purchase** - 17,622,107 reviews  
3. **Wireless** - 9,001,567 reviews
4. **PC** - 6,908,245 reviews
5. **Home** - 6,221,321 reviews
6. **Apparel** - 5,906,085 reviews
7. **Health & Personal Care** - 5,331,239 reviews
8. **Beauty** - 5,115,462 reviews
9. **Video DVD** - 5,069,014 reviews
10. **Mobile_Apps** - 5,033,164 reviews

It's interesting to note that Books and Digital Ebook Purchase (which are related categories) together account for over 37 million reviews, showing the strong popularity of reading materials on Amazon's platform.
================================================================================

​Requisitos previos

Requisitos previos