THE CHALLENGE
Claude’s careful. It reads the tables it can reach, names the column it used and explains its work. But it’s still working from a definition picked by a model that couldn’t look yours up.
PARTNERS / ANTHROPIC
AtScale and Claude: ask anything, get the number Finance filed.
AtScale connects to Claude, Claude Code and Cowork over MCP. Ask in your own words, and every number you get back tells you which definition it came from.
WHAT YOU GET
Claude reads your metrics, dimensions, hierarchies and access rules from one semantic model, checks which metric you meant when a word has more than one, and returns a number ACE computed.
SEE IT IN CLAUDE
Ask about churn, and Claude checks which churn you mean.
AtScale Ava (Ask, Verify, Answer) finds every metric in the model that answers to the word churn, lays them out with their owners and hands the choice to you before anything gets computed. You pick. Then ACE (the AI Computation Engine) does the math.
You ask Claude
"How bad was churn last year?"
↓ ATSCALE AVA CHECKS THE MODEL
AtScale Ava
Your model has five metrics that match "churn." Which one did you mean?
1. Logo churn, owned by RevOps
2. Gross revenue churn
3. Net revenue retention
4. Voluntary churn
5. Involuntary churn
The board deck uses 1.↓ ATSCALE ACE COMPUTES IT
1.8%
Q2 FY26 · logo churn
Computed by AtScale ACE · definition: churn_logo · owner: RevOps · suspended accounts excluded
WHAT IT DOES
One server, three places Claude works.
AtScale runs a Model Context Protocol server. Claude connects to it and reads the metrics, dimensions, hierarchies, relationships and access rules you’ve already defined.
The same server works for the analyst in Claude, the engineer building against the metrics in Claude Code and the person in Cowork who wants a recurring report done. You set it up once, and all three use the same model.
HOW IT CONNECTS
You connect it once, and Claude can use it everywhere.
01
Add the server
AtScale’s MCP server goes into the connector settings once for the workspace.
02
Sign in as yourself
Claude inherits your identity, so the model’s access rules apply to what you see.
03
Ask in your own words
Claude sees the metrics as tools, and Ava checks with you when a word maps to more than one.
04
Read the answer with its definition
Every figure names its metric and owner, so it’s safe to quote in a board deck.
MORE WAYS TO USE IT
What teams do once Claude can reach the metrics.
Analysis without a ticket
Analysts reach the metrics directly, so the data team isn’t the queue anymore.
Commentary that matches the number
Claude writes the narrative around a figure ACE computed.
Agents on the Claude API
Whatever you build reads the same metrics your analysts read.
Reports finished in Cowork
A recurring report gets produced end to end.
Model building with CoModel
Claude Code drafts model changes and a person approves them. CoModel arrives this fall.
Answers tested by Canary
Answers get tested against known-correct results before anyone acts on them. Canary arrives this fall.
PROOF
Teams already running their metrics on AtScale.
Case StudiesBlue Yonder
Case StudiesPapa Johns
Case StudiesCloud-Based OLAP Transformation: How Vodafone Portugal, Celfocus, and AtScale Modernized Analytics
FAQ
Questions we hear.
Which Claude products does AtScale work with?
Claude, Claude Code and Cowork, all through AtScale’s MCP server. You add the server once and all three use the same model. Agents you build on the Claude API can call the same metrics too.
Where is the number computed?
In your own data platform. ACE (the AI Computation Engine) computes the metric where the data already sits, so your data doesn’t move. Claude gets the result along with the definition and owner behind it.
What if my prompt is vague?
Ava checks which metric you meant. It finds every metric that matches the word you used, shows them with their owners and lets you pick one before ACE computes anything.
Can Claude see data I can't?
No. You sign in as yourself with OAuth, so the model’s access rules apply to Claude the same way they apply to you. The MCP tools are read-only.
When are CoModel and Canary available?
Both are coming this fall. CoModel will let Claude Code draft model changes for a person to approve, and Canary will test answers against known-correct results before anyone acts on them.
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.