---
name: mprove-connection-schemas
description: Get, create or update schema metadata for Mprove project connections
---

# mprove-connection-schemas

Call `read-docs` tool with `pageId: dwh-schema`

## Schema Files

Schema files provide metadata (descriptions, examples, relationships, cache_unique_values) for database tables and columns.
They are combined with raw database schemas to build "combined schemas" that are useful for creating Malloy semantic models.

## When to Use

Call `get-connections-list` and `get-schemas` tools when: 
- User requested to create or edit Mprove model based on Malloy

Call `get-connections-list`, `get-schemas` tools and create/update `.schema` files when: 
- User asks to setup Mprove project
- Schema file does not exist for Mprove project SQL connection
- User requested to create or update schema metadata for a SQL connection

## File Location and Naming

- Directory: `schemas/` inside the `mprove_dir` directory (e.g., `data/schemas/`)
- Naming: `<connection_name>.<schema_name>.schema`
- Example: `c1_postgres.fleet.schema`

## Workflow for create/update .schema files

### Step 1: Fetch Combined Schemas

Read `/home/user/.config/opencode/mprove-session.json` (if file exist) to get `projectId`, `repoId`, `branchId`, and `envId`.

Call `get-connections-list` and `get-schemas` MCP tools.

### Step 2: Check Existing Schema Files

Look for existing `.schema` files in `data/schemas/` if need to create/update metadata.

### Step 3: Write Schema Files

For tables of interest, call the `get-sample` MCP tool.

Use `foreignKeys` and sample data to infer relationships and to understand if we should cache unique values for the column. 

For non-sensitive data write realistic examples.

Create or update `.schema` files in `data/schemas/`. 

Guidelines:
- Include all tables and columns
- Write clear, business-oriented descriptions
- For non-sensitive data set `example` value
- Speficy `cache_unique_values: true` to recommend column unique values to be cached. 
Cache will help to translate user data questions to filters of query.
Good candidates are string columns that are non-id. 
Set `cache_unique_values: true` for columns like categories, status, company name, etc.