Automation · Operations and reporting
Ask on WhatsApp how much was sold and answer with a ClickHouse query
You ask on WhatsApp how much was sold and an AI agent finds the table in ClickHouse Cloud, runs a SELECT and answers with the figure.
How do I ask my ClickHouse data on WhatsApp how much we sold yesterday?
You write to the Met on WhatsApp "how much did we sell yesterday by city?". The Met finds your organization and the ClickHouse Cloud service that holds the data, lists its databases and reads the columns of the sales table so it does not guess names. Then it runs a SELECT query that groups by city and filters by date, and answers with the figures, without pasting the whole table. Everything is read only: the ClickHouse server only accepts SELECT, so the Met cannot change or delete anything even if asked.
What starts it
When someone asks for it
It starts when someone on the team writes to the Met on WhatsApp with a question about sales, orders or any data that lives in ClickHouse. There is no trigger to configure: the Met answers inside that conversation.
A manual task: someone on the team runs it with a button, or asks the Met for it in chat.
Who runs it
Analyst Met that queries ClickHouse
This is not a named, seeded Met. It is an internal Met, with ClickHouse Cloud connected, that answers the owner or the team on WhatsApp: its work ends when the person has the figure they asked for, taken from a query and not from memory.
Channels: WhatsApp
What you need connected
- ClickHouse Cloud Research and analyze
5 steps · 5 tools
The procedure, step by step
Each step shows the tool that runs. They belong to the real connector: if one of them stopped existing, this page would not build.
- 1
Find the organization and the service
The Met gets the organizations your user can reach and lists their services. It keeps the one you named in the skill, which must have MCP enabled; if it does not, it cannot query it.
- 2
Find the database that holds sales
The Met lists the service's databases and picks the right one. If several have similar names, it asks which one before going on.
- 3
Read the table's columns
The Met lists the database's tables with their column definitions, filtering by name if needed, to know which one stores the date, the amount and the city. That way it does not make up a column.
- 4
Run the SELECT query
With the real columns, the Met writes a SELECT query that filters by the requested date and groups by city, and runs it on the service. The server rejects anything that is not a SELECT.
- 5
Answer with the figure in the chat
The Met answers with the totals by city and the grand total, and says which table and which date they come from. If the person wants another breakdown, such as by product, it queries again.
Before you start
What you need ready
The Met solves none of these for you. If one is missing, the automation stops there.
-
ClickHouse Cloud with MCP enabled
In the ClickHouse Cloud console, open the service, go to Connect, choose Connect with MCP and turn the toggle on. Then, in Met, open Connections → ClickHouse Cloud → Connect.
-
A sales table with clear columns
The Met reads the columns before querying. If the date, the amount and the city have clear names, it gets the query right the first time.
-
An internal Met on WhatsApp
Assign a Met with the ClickHouse skill to the team's WhatsApp line, and write in the skill which service and which database to use.
The limits
What this automation does NOT solve
Better said here than discovered halfway through the rollout.
-
It does not load or fix data: the server only accepts SELECT. If a sale is not in the table, the Met does not see it.
-
It does not guess what "selling" means in your business. If the table has voided sales or returns, write in the skill how to filter them so the query excludes them.
-
It does not query services where you did not enable MCP, even if your user sees them in the console.
-
It does not replace a dashboard: it answers one question at a time in the chat. To track a figure every day you need another tool.
FAQ
Questions about this automation
No. ClickHouse says run_select_query only allows SELECT and that all queries are read only, so a delete order is rejected.
Yes. The Met runs the query when you ask and keeps no copy, so it answers with what is in ClickHouse at that moment.
Yes, as long as it is in a service with MCP enabled. The Met lists the tables and reads their columns before querying it.
You do, with your ClickHouse Cloud account: queries run on your services. Meteor does not resell ClickHouse.
Keep going
Other automations built on the same tools
Operations and reporting
Turn off on WhatsApp the feature flag that breaks the checkout in LaunchDarkly
How do I quickly turn off a feature flag that broke the checkout without opening LaunchDarkly?
- Starts:
- when someone asks for it
- Steps:
- 5, with 7 real tools
- Needs:
- LaunchDarkly
Operations and reporting
Build a department's higher education report without opening SNIES
How do I pull SNIES enrollment and program figures without going into the portal?
- Starts:
- when the customer writes
- Steps:
- 5, with 5 real tools
- Needs:
- SNIES
Operations and reporting
Build your team's org chart in Lucid on WhatsApp and share it
How do I build my business's org chart without sitting down to draw it?
- Starts:
- when someone asks for it
- Steps:
- 6, with 6 real tools
- Needs:
- Lucid
Operations and reporting
Tell customers on your status page that the store is down, asking on WhatsApp with Better Stack
How do I tell my customers my online store is down without opening Better Stack?
- Starts:
- when someone asks for it
- Steps:
- 6, with 6 real tools
- Needs:
- Better Stack
Want this automation running on your own data?
We will show it to you with your accounts connected, not with a canned demo.