Using ClickHouse MCP server with Open WebUI

This guide explains how to set up Open WebUI with a ClickHouse MCP server and connect it to the ClickHouse example datasets.

Install uv

You will need to install uv to follow the instructions in this guide. If you don't want to use uv, you will need to update the MCP Server config to use an alternative package manager.

Launch Open WebUI

To launch Open WebUI, you can run the following command:

uv run --with open-webui open-webui serve

Navigate to http://localhost:8080/ to see the UI.

Configure ClickHouse MCP Server

To setup the ClickHouse MCP Server, we'll need to convert the MCP Server to Open API endpoints. Let's first set environmental variables that will let us connect to the ClickHouse SQL Playground:

export CLICKHOUSE_HOST="sql-clickhouse.clickhouse.com"
export CLICKHOUSE_USER="demo"
export CLICKHOUSE_PASSWORD=""

And, then, we can run mcpo to create the Open API endpoints:

uvx mcpo --port 8000 -- uv run --with mcp-clickhouse --python 3.10 mcp-clickhouse

You can see a list of the endpoints created by navigating to http://localhost:8000/docs

To use these endpoints with Open WebUI, we need to navigate to settings:

Click on Tools:

Add http://localhost:8000 as the tool URL:

Once we've done this, we should see a 1 next to the tool icon on the chat bar:

If we click on the tool icon, we can then list the available tools:

Configure OpenAI

By default, Open WebUI works with Ollama models, but we can add OpenAI compatible endpoints as well. These are configured via the settings menu, but this time we need to click on the Connections tab: