Как создать агента ИИ с помощью Agno и MCP-сервера ClickHouse

В этом руководстве вы узнаете, как создать агента ИИ на базе Agno, способного взаимодействовать с SQL-песочницей ClickHouse, используя MCP-сервер ClickHouse.

Пример ноутбука

Этот пример доступен в виде ноутбука в репозитории с примерами.

Предварительные требования

• В вашей системе должен быть установлен Python.
• В вашей системе должен быть установлен pip.
• Вам потребуется API-ключ Anthropic или API-ключ другого поставщика LLM.

Следующие шаги можно выполнить либо из Python REPL, либо с помощью скрипта.

Установка библиотек

Установите библиотеку Agno, выполнив следующие команды:

pip install -q --upgrade pip
pip install -q agno
pip install -q ipywidgets

Настройка учётных данных

Далее необходимо указать ваш API-ключ Anthropic:

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

Enter Anthropic API Key: ········

Использование другого LLM-провайдера

Если у вас нет API-ключа Anthropic и вы хотите использовать другой LLM-провайдер, инструкции по настройке учётных данных можно найти в документации Agno

Далее определите учётные данные, необходимые для подключения к ClickHouse SQL playground:

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

Инициализация MCP-сервера и агента Agno

Теперь настройте ClickHouse MCP-сервер для подключения к ClickHouse SQL playground и инициализируйте агента Agno, задав ему вопрос:

from agno.agent import Agent
from agno.tools.mcp import MCPTools
from agno.models.anthropic import Claude

async with MCPTools(command="uv run --with mcp-clickhouse --python 3.13 mcp-clickhouse", env=env, timeout_seconds=60) as mcp_tools:
    agent = Agent(
        model=Claude(id="claude-3-5-sonnet-20240620"),
        markdown=True,
        tools = [mcp_tools]
    )
await agent.aprint_response("What's the most starred project in 2025?", stream=True)

▰▱▱▱▱▱▱ Thinking...
┏━ Message ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃                                                                                                                 ┃
┃ What's the most starred project in 2025?                                                                        ┃
┃                                                                                                                 ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
┏━ Tool Calls ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃                                                                                                                 ┃
┃ • list_tables(database=github, like=%)                                                                          ┃
┃ • run_select_query(query=SELECT                                                                                 ┃
┃     repo_name,                                                                                                  ┃
┃     SUM(count) AS stars_2025                                                                                    ┃
┃ FROM github.repo_events_per_day                                                                                 ┃
┃ WHERE event_type = 'WatchEvent'                                                                                 ┃
┃     AND created_at >= '2025-01-01'                                                                              ┃
┃     AND created_at < '2026-01-01'                                                                               ┃
┃ GROUP BY repo_name                                                                                              ┃
┃ ORDER BY stars_2025 DESC                                                                                        ┃
┃ LIMIT 1)                                                                                                        ┃
┃                                                                                                                 ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
┏━ Response (34.9s) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃                                                                                                                 ┃
┃ To answer your question about the most starred project in 2025, I'll need to query the ClickHouse database.     ┃
┃ However, before I can do that, I need to gather some information and make sure we're looking at the right data. ┃
┃ Let me check the available databases and tables first.Thank you for providing the list of databases. I can see  ┃
┃ that there's a "github" database, which is likely to contain the information we're looking for. Let's check the ┃
┃ tables in this database.Now that we have information about the tables in the github database, we can query the  ┃
┃ relevant data to answer your question about the most starred project in 2025. We'll use the repo_events_per_day ┃
┃ table, which contains daily event counts for each repository, including star events (WatchEvents).              ┃
┃                                                                                                                 ┃
┃ Let's create a query to find the most starred project in 2025:Based on the query results, I can answer your     ┃
┃ question about the most starred project in 2025:                                                                ┃
┃                                                                                                                 ┃
┃ The most starred project in 2025 was deepseek-ai/DeepSeek-R1, which received 84,962 stars during that year.     ┃
┃                                                                                                                 ┃
┃ This project, DeepSeek-R1, appears to be an AI-related repository from the DeepSeek AI organization. It gained  ┃
┃ significant attention and popularity among the GitHub community in 2025, earning the highest number of stars    ┃
┃ for any project during that year.                                                                               ┃
┃                                                                                                                 ┃
┃ It's worth noting that this data is based on the GitHub events recorded in the database, and it represents the  ┃
┃ stars (WatchEvents) accumulated specifically during the year 2025. The total number of stars for this project   ┃
┃ might be higher if we consider its entire lifespan.                                                             ┃
┃                                                                                                                 ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛