MCP trust layer for metric agents
AI infrastructure for autonomous data and metric agents.
Your AI keeps forgetting the metric logic your team already approved.
ChatData saves the approved definition, source, caveats, proof receipt, and rerun path for recurring KPI questions, then gives Claude Code, Cursor, and Codex the same trusted route next week.
Free for 7 days. Decide what to keep, expand, or stop on a day-7 live call.
The recurring metric loop
The answer exists. The next run still starts over.
Teams keep metric truth in Notion, dbt catalog pages, dashboard screenshots, shared Sheets, and analyst memory. Claude, Cursor, and Codex can answer fast, but the number is not worth sharing until they use the reviewed route.
Metric truth is scattered
The definition lives in one place, the SQL lives in another, and the dashboard owner remembers the caveat. Claude sees the metric name and fills in the rest.
Proof is hard to repeat
A dashboard screenshot helps today. A dbt model helps if the agent can read it. Neither helps next week unless the source, freshness, caveat, and validation rule are saved together.
Fixes don’t compound
You correct the same answer in the thread, deck, or notebook. Nothing carries into the next Claude, Cursor, or Codex run, so the team pays the same cleanup cost again.
How it works
Six checks run before a saved answer can be reused.
Existing data catalogs were built for humans to browse. ChatData gives Claude Code, Cursor, and Codex the reviewed route for one metric answer: definition, source, freshness, caveat, validation, and proof.
Under the hood, ChatData can store metric definitions using the open Open Semantic Interchange core spec.
Right metric definition?
Resolves against your governed metric registry, not what the agent guessed from the schema.
Source trusted and fresh?
Checks the approved source and last-refreshed timestamp before any answer is formed.
Frame stated and tested?
Names the explanatory frame, what supports it, and what would break it before committing.
Data passed quality checks?
Uses saved validation rules, expected ranges, and source tie-outs before drawing conclusions.
Caveats visible?
Known exceptions, lags, and schema quirks are attached from saved context.
Rerun works next time?
After review, MCP pulls the saved route again. Invite only after the rerun uses the same metric, source, caveat, and proof.
Wrong grain. No caveat. Sounds right.
Notion, dbt, Sheets, dashboard screenshot, old thread.
Recalculate, add caveats, re-share.
Nothing was saved. Repeat.
Governed definition, approved source, correct grain.
iOS lag, schema note, freshness — pulled from saved context.
Source, caveat, proof receipt, and validation path are visible.
Next MCP pull starts from the reviewed route.
Trial access
Bring one metric question your team keeps re-answering.
Start with a dashboard screenshot, dashboard URL, GitHub SQL/dbt link, metric definition, or source doc. No card. Trial access opens after work-email verification.