The Atlan MCP Cookbook

Build a new table or view on trusted sources

Before you write the DDL, use Atlan to pick the right sources, confirm the grain and definitions, and avoid rebuilding something that already exists.

All recipes

On the menu

Use case
Find & understand data
Action
Look up
Persona
AI Builder, Data Engineer
Effort
Moderate
Runs in
any AI tool

The task

Scenario. You've been asked to build a new customer_360 view. The risk isn't the SQL, it's building on the wrong sources, at the wrong grain, or re-deriving a metric the company already has. Before you write a line of DDL, use Atlan to find the certified sources to build on, confirm what the key fields mean, and check that nothing equivalent already exists.

The ask

  • The certified source tables to build on
  • The governed definitions and grain of the key fields
  • Whether an equivalent table or view already exists

Hero prompt

Hero prompt

I'm about to build a customer_360 view. Which certified tables should I build on, what are the governed definitions of the key fields, and does an equivalent view already exist?

Step by step

  1. Find the certified sources to build on, and flag lookalikes to avoid.

    Step 1

    I'm building a <new table/view> for <purpose> — which certified tables should I build on (with grain and owner), and which lookalikes should I avoid?

  2. Confirm the governed definitions of the key fields.

    Step 2

    Show the governed definitions of <key fields> so I model them correctly.

  3. Check whether an equivalent already exists.

    Step 3

    Does an equivalent table or view already exist that I should reuse instead of rebuilding?

What you'll give it

Required context

  • What you're building and its purpose
  • Lineage and certification populated for the candidate sources

Optional context

  • The metrics or fields it must include, so definitions can be confirmed
  • Your target schema or layer, so recommendations fit your stack

What Atlan creates

Outputs

  • The certified source tables to build on (grain, owner, certification), and the lookalikes to avoid
  • The governed definitions of the key fields
  • Any existing table or view that already covers this, so you reuse instead of rebuild

Summary. You start the DDL from trusted sources, at the right grain, with the official definitions, and you don't rebuild what already exists.

Read-only

Nothing is written. It reads sources, lineage, and definitions to inform the build; you write the table or view in your warehouse.

Follow-up prompts

Keep going once the first answer lands.

Map the joins

Map the joins

For those source tables, what join keys and filters do people already use?

Draft the outline

Draft the outline

Draft a column list for the new view with the governed definition next to each field.

Sanity-check the grain

Sanity-check the grain

At <grain>, which source is the driving table and which are lookups?

Tips & troubleshooting

Cook's note

The expensive mistake isn't a slow query, it's building on the wrong source or re-deriving a metric that already exists. Five minutes here saves that.

  • Match the tool to the work. Run it in whatever AI tool you build in; it reads, so it's safe to run before you write any DDL.
  • Automate or chain it. Chain it to your modeling workflow so every new model starts from certified sources and governed definitions.
  • It only reads. Nothing is written to Atlan; you build the table or view in your warehouse from the answer.
THE ATLAN MCP COOKBOOK

Point your AI at the Atlan MCP.
Run your first recipe.

[Website env: production]