Context layer for your data warehouse

AI gives confident answers.
But are they actually right?

Contextary teaches Claude, Cursor, and any AI tool what your data actually means: metric definitions, gotchas, business rules. So the answers it gives are right.

Start for free See how it works

Free to start · No credit card · 60-second setup

Claude contextary

You

What's our Q3 pipeline value?

get_table_context("opportunities")

amounts in cents stage 0 ≠ qualified exclude test accounts fiscal Q3 = May-Jul

Q3 qualified pipeline is $3.2M across 47 open opportunities.

Verified against your warehouse without context: ~~$12.4M~~

The cost of missing context

Confidently wrong is worse than wrong.

Without your team's knowledge, even the best AI produces answers that look perfect and quietly aren't.

Pipeline forecast

AI said ~~$12.4M~~

$3.2M

Stage 0 counted as sales-qualified

AI counts every open opportunity as pipeline. Your VP walks into Monday's forecast call with a number that's 4× reality.

MQLs this quarter

Marketing said ~~10K~~

A different definition every time

Marketing reports 10K MQLs. Sales says 6K. AI used a different definition of "qualified" each time. Nobody knows which is right.

Churn, board deck

AI said 8%

Trials and paused accounts included

Trial accounts land in the churn calculation. Paused customers show as churned. The board sees double the real number.

Set up in 60 seconds, not 6 sprints

Two clicks here.
One line in Claude.
Done.

No data warehouse migration. No new modeling layer. No engineering quarter blocked on integration. Connect your warehouse, paste an MCP URL into Claude, start asking.

01 Connect your warehouse: Snowflake, BigQuery, Postgres, or Redshift. Contextary auto-discovers your tables and columns.
02 Annotate the parts that matter. AI suggests starter annotations from your schema; you accept, edit, or rewrite.
03 Paste the MCP URL into Claude (or Cursor, or Claude Code). Every AI tool now answers from your handbook.

claude_desktop_config.json Copy

// One line, in Claude's MCP config:

  "mcpServers": 
    "contextary": 
      "url": "https://app.contextary.ai/mcp"
    
  


// That's it. Claude now answers from your handbook.

What's inside

One source of truth for what your data means.

Document your team's knowledge once. Every AI tool, dashboard, and teammate uses the same definitions.

metrics

Metrics everyone agrees on

Define MRR, churn, NRR once. Every AI tool, dashboard, and teammate uses the same formula. No more "which number is right?"

gotchas

Gotchas that catch mistakes first

Flag the tricky stuff: cents vs. dollars, soft deletes, test accounts. AI never trips on them and your reports stay accurate.

verify

Verify before you ship

Test AI-generated SQL against real data. Catch wrong results before they hit a dashboard or board deck.

mcp

Works with every AI tool

Claude Desktop, Cursor, Claude Code, anything that speaks MCP. Set up once, works everywhere your team writes prompts.

connectors

Connect any warehouse

BigQuery, Snowflake, Postgres, Redshift. Schema metadata only. Your data never leaves your warehouse.

bi sync

Sync back to BI

Annotations push back as table and column comments. Looker, Tableau, and dbt inherit the context for free.

The philosophy

Let AI do what it's great at. Just give it the right context.

AI is already incredible at reasoning, writing SQL, and building dashboards. The only thing it's missing is knowing what your data actually means.

AI is already amazing at

✓ Writing complex SQL across multiple tables
✓ Building interactive dashboards and reports
✓ Creating charts and data visualizations
✓ Summarizing trends and spotting anomalies
✓ Explaining data in plain English for stakeholders

Contextary fills in the gaps

✓ Which join keys to use (and which to avoid)
✓ What "revenue" actually means in your company
✓ Data gotchas like cents vs. dollars and soft deletes
✓ Your team's metric definitions and business rules
✓ Where data comes from and what depends on it

“Claude can already build you a dashboard in seconds. Contextary just makes sure the numbers on it are right.”

Works with your stack.

Connects to your warehouse and serves context to the AI tools your team already uses.

AI tools

Claude Desktop · Cursor · Claude Code · anything that speaks MCP

Warehouses

BigQuery · Snowflake · PostgreSQL · Redshift

Coming up

dbt · Looker · Tableau

AI gives confident answers. But are they actually right?