THE CHALLENGE
A coding agent reads the schema, finds a column called revenue, joins it to something plausible and hands back SQL that runs. Working SQL looks the same whether the definition behind it is yours or improvised, so a reviewer can’t tell.
PARTNERS / SPACEX Super Intelligence / GROK
AtScale and SpaceX Super Intelligence: coding agents that use Finance’s definitions.
AtScale has joined the SpaceXSI Partner Network as a Marketplace Partner. With the AtScale plugin for Grok Build, Grok connects to AtScale over MCP and asks for the metrics Finance has signed off on. You keep the same agent and the same speed, and the figure it returns matches the finance report.
WHAT YOU GET
Grok reads metrics, dimensions and hierarchies from your semantic model, checks which metric you meant and gets back a number AtScale ACE computed from the definition Finance uses.
SEE IT IN THE TERMINAL
Your model has five kinds of revenue. Grok checks which one you need.
Without that step, the question returns one number and no sign the other four existed. The answer looks finished, so nobody’s likely to check it.
WHAT IT DOES
Point Grok at the model, and let ACE handle the warehouse.
The tenure rule, the exclusion list, the fiscal calendar and the reason Finance uses a different column aren’t in the schema. An agent that’s reading tables has to guess at every one of them.
With AtScale, it doesn’t have to. Grok connects to AtScale’s Model Context Protocol (MCP) server and reads the metrics, dimensions, hierarchies, relationships and access rules you’ve defined. ACE (the AI Computation Engine) computes each metric where the data already sits, so it’s the same value Finance published.
HOW IT CONNECTS
It takes four steps to set up in Grok.
MORE WAYS TO USE IT
What developers build on top of it.
Models as code
A definition change arrives as a pull request instead of a ticket.
Metrics in your own app
Your product reads the same numbers your finance team reports.
Generated SQL that matches hand-written SQL
The agent and the analyst use one definition.
Grok Bots that use the same metrics
A Bot your sales team runs reads the same definitions your code does.
Test a change before it ships
Run the old definition beside the new one and see which reports move.
An audit trail that’s just git
Every change to a metric has an author, a review and a date.
PROOF
Who runs AtScale today.
Case StudiesBlue Yonder
Case StudiesPapa Johns
Case StudiesCloud-Based OLAP Transformation: How Vodafone Portugal, Celfocus, and AtScale Modernized Analytics
FAQ
Questions we hear.
How does Grok Build connect to AtScale?
Through AtScale’s MCP server. You add it to Grok Build’s MCP configuration and sign in as yourself with OAuth, so your access rules apply to the agent the same way they apply to you. The MCP tools are read-only.
Does it work in Cursor too?
Yes. The same MCP server works in Cursor, so teams moving to Grok Build keep the same setup.
Does the agent still write SQL?
Not for metrics. ACE (the AI Computation Engine) computes each metric from the definition in your model and runs the SQL itself. You can read the query it used.
What if my prompt could mean more than one metric?
Ava checks which one you meant. If your model has five kinds of revenue, Ava finds all five and confirms the one you need, like recognized revenue for the quarterly report, before ACE computes it.
Which data platforms does it work with?
Snowflake, Databricks, Google BigQuery, Amazon Redshift, PostgreSQL and InterSystems IRIS. Self-hosted AtScale also runs on Microsoft SQL Server, Azure SQL, Azure Synapse, Cloudera CDP and IBM Db2. ACE computes on the platform you already run, so your data doesn’t move.
Which other AI tools can connect?
Any client that supports remote MCP servers. That includes Claude, ChatGPT, Google Gemini and Databricks agents alongside Grok Build, and they all request the same metric definitions.
Book a workshop.
- Bring us your most controversial metric and the business questions your team and your agents ask about it every week.
- We benchmark your AI on them, with AtScale and without it.
- You keep the semantic model we built to run the test.