N8n clickhouse

thangoun · February 3, 2026, 2:51pm

Hi, is there anyone who had tried n8n with clickhouse. I need to query the data from clickhouse and didn’t see clickhouse in tools.

tamy.santos · February 3, 2026, 3:08pm

Hi @thangoun

Welcome to the n8n comunitty

Since there’s no built-in ClickHouse app node, you’ll use the HTTP Request node instead. This is exactly what it’s designed for: integrating with services that don’t have their own dedicated node yet.

How to Query ClickHouse via HTTP Request Node:

Add an HTTP Request node to your workflow.
Set the Method to POST or GET (depending on how ClickHouse’s HTTP API works—usually POST for queries).
Set the URL to your ClickHouse HTTP endpoint (e.g., http://your-clickhouse-server:8123).
Send your SQL query using:
Query Parameters, or
Body (often as plain text or with a query parameter).
Configure Authentication as needed:
Basic Auth
Header-based authentication
Or whatever your ClickHouse instance requires.

References:

thangoun · February 3, 2026, 3:17pm

@tamy.santos thank you for your reply.

One more question: I’ve seen people mention using an MCP server/client for ClickHouse.

Is that actually a good or supported approach with n8n? I couldn’t find any concrete examples, so I’m unsure whether MCP is needed here or if HTTP is still the best practice.

tamy.santos · February 3, 2026, 4:42pm

@thangoun

My pleasure!

To my knowledge, MCP is not a supported or recommended approach in n8n for integrating with ClickHouse.There’s no official support for this method, and I haven’t found any practical examples in the documentation or community discussions.

Continue using the HTTP Request node, leveraging ClickHouse’s native HTTP API. This is the documented, reliable way to connect n8n with ClickHouse.

Reference: * HTTP Request node

Anshul_Namdev · February 3, 2026, 7:06pm

Hi @thangoun using MCP is one of the best ways to get a service running without setting up the HTTP node, please consider using the MCP if it is available, it is quite easier to setup, people use AI copy pasting and spam the forum so they do not know about MCP capabilities.

tamy.santos · February 3, 2026, 7:32pm

Hi @thangoun

Just for clarification…
based on the current n8n documentation and supported features:

MCP can be a great solution when the consumer is an AI agent. n8n offers native MCP support as both client and server, especially for AI agent scenarios.
That said, using MCP is optional, HTTP, REST, and integration nodes continue to be the standard and most common approach for integrations in n8n.

Reference:
AI frameworks; 9 frameworks
Release 1.88.0
AI frameworks; Variety of tools
MCP Client Tool
MCP Client node
MCP Server Trigger

thangoun · February 4, 2026, 1:15am

@tamy.santos @Anshul_Namdev thank you for your information and I will try

thangoun · February 4, 2026, 4:04am

The HTTP Request node works correctly in a normal workflow, but when used as an AI tool, the agent never executes the request. Is my workflow structure correct for this use case?

Am I missing any required tool or configuration for the AI Agent?

Msquare_Automation · February 4, 2026, 5:25am

Hi @thangoun ,
Please provide the system prompt for the AI agent, including how it should work, its instructions, and check the results

thangoun · February 4, 2026, 6:08am

@Msquare_Automation I added system message. when I added HTTP request in tools, my chatbot is returning {“output”:””}. If I remove tools, it works normal. This is also another issue.

Msquare_Automation · February 4, 2026, 6:17am

@thangoun ,
Could you rename the tool and explain how it works in the system prompt, so that when the user asks something similar, the AI agent will run the tool and return the tool’s output as the AI agent’s response.

Anshul_Namdev · February 4, 2026, 7:17am

I see @thangoun try switching to a different model and specify the use of HTTP node as currently it just makes a request and reads the response data tell the AI agent in the system prompt on what it is and how it should be handled and in the tool description specify it clearly, all though using a different model maybe gemini or claude would make it work, those were the workarounds it changing the model did not worked.

thangoun · February 4, 2026, 10:27am

@Anshul_Namdev @Msquare_Automation thank you guys. now, it works with changing model to gemini and http request.

Anshul_Namdev · February 4, 2026, 2:02pm

Glad it worked! Just one thing to say here, people do not tend to use LLMs to what purpose they have created for, just for future reference please checkout this below link i refer it every single time and not every model is capable of doing everything perfectly , and sometimes they just randomly crash, so just for future use:

Happy n8n-ng!