# MCP with verified SQL reports and reviewable AI context | DRE

> Use DRE MCP for reviewed report answers, local refinements and documented table exploration. Review provenance, rendered SQL and exact downloads.

[← All use cases](https://getdre.com/use-cases/)

# Trustworthy AI with MCP

[Request MCP access →](mailto:hello@getdre.com?subject=MCP%20access%20request)

Most of the trustworthy context already lives in recurring reports. DRE brings reviewed report logic and upstream pipeline comments into a MCP server, so an agent can find an answer, show its origin and let you review the rendered SQL.

Local RC demo host using real MCP tool results: verified workbook download, local aggregation and an exploratory join with rendered SQL.

Local RC demo host using real MCP tool results: verified workbook download, local aggregation and an exploratory join with rendered SQL.

reports/revenue/revenue.yml

```yaml
description: Net revenue by customer region, excluding GST
meta:
  units: AUD
  ai:
    # Exposure is reviewed and enabled by the owner.
    expose: true
    questions:
      - What is net revenue by region?
      - Which region has the most revenue?
queries: [revenue]
output: {format: xlsx}
```

reports/revenue/revenue.sql

```sql
select c.region, sum(o.net_amount) as revenue
from {{ source('shop', 'orders') }} o
join {{ source('shop', 'customers') }} c
  on o.customer_id = c.customer_id
group by c.region
order by c.region
```

.dre-mcp.yml

```yaml
# Report-linked tables are exposed by default.
expose_report_tables: true
# Additional tables and schemas require explicit opt-in.
# tables:
#   warehouse: [analytics.customer_targets]
# Exclusions override inherited exposure.
# exclude_tables:
#   warehouse: [analytics.customer_private]
```

terminal

```bash
# Requires DRE 0.4.1 or later.
dre extension install dre-mcp@0.1.0-rc.1 --accept-proprietary
# Read the complete EULA before accepting it.
dre mcp license show
dre mcp license accept --version 2026-10
dre mcp validate
dre mcp test
dre mcp serve
```

### What to notice

-   Verified report: run reviewed SQL and download the exact report output with its variables, time and SQL hashes.
-   Derived from verified result: ask DRE to aggregate or filter saved rows through its installed DuckDB plugin, retaining the source and transformation.
-   Exploratory (unverified): join explicitly exposed, documented tables. A known report relationship is context, not proof of a new answer.
-   Review compiled or generated SQL in the collapsed card section, or download it. Refresh context without restarting the server.
-   DRE works well with a good dbt setup or data pipeline. Table and column comments come from that pipeline; DRE adds recurring report logic.
-   Report-related tables are exposed by default. Disable with --no-report-tables or expose\_report\_tables: false; extra tables and schemas stay opt-in.
-   MCP is a thin companion. DRE and its plugins handle database connections and execution. Install the DuckDB plugin for saved-result refinements, even when reports use another database.
-   MCP is in early RC testing. Request access and tell us how you would use it.

### In the docs

-   [MCP documentation →](https://getdre.com/docs/mcp/)
-   [Prepare AI-ready reports →](https://getdre.com/docs/ai-ready-reports/)

## More use cases

[

### Monthly finance pack

A multi-sheet Excel workbook with money formats, formulas, totals and a branded template.

](https://getdre.com/use-cases/monthly-finance-pack/)[

### A file for a bank or regulator

Fixed-width or delimited files to the exact layout the other side asks for, dropped on SFTP.

](https://getdre.com/use-cases/file-for-a-bank/)[

### Extracts to S3, GCS or Azure Blob

Parquet or csv extracts, written where downstream systems pick them up.

](https://getdre.com/use-cases/extracts-to-object-storage/)