Case study

Putting Supabase data to work with Modiva

A look at how teams use Modiva with Supabase — querying Postgres, reading projects, edge functions and storage across their stack, and giving AI agents safe, scoped, revocable access — without building and maintaining a custom integration.

This is a representative example of the Modiva + Supabase workflow. Have a story to share? Tell us and we’ll feature it here.

The challenge

Teams that build on Supabase rarely want every data question to end at the dashboard. Getting that data where the work happens usually means:

• Answering product questions meant a developer opening the Supabase dashboard or a psql session to run the query by hand.
• Internal tools that read production data each carried their own service-role keys and hand-rolled Management API clients.
• AI assistants had no scoped, revocable way to inspect projects or query the database — sharing a raw connection string was the only option.

The approach with Modiva

1. 1

   One connection, not a shared connection string

   The team authorizes Supabase in Modiva once via the Management API OAuth flow (PKCE). Modiva stores only a scoped, encrypted token and refreshes it automatically — no service-role keys in scripts, no connection strings passed around.

2. 2

   SQL and project management over a unified API

   Instead of writing against the Management API directly, the team calls Modiva's typed tools — run a SQL query, list tables, create projects, deploy edge functions, manage secrets and branches. The same request shape works across every tool Modiva supports, so adding the next system is a connection, not a rewrite.

3. 3

   Database access for AI agents via MCP

   An MCP key exposes the same tools to AI assistants. An agent can list projects, inspect the schema and answer a data question with a query — with access the team controls and can revoke at any time. The SQL tool executes what it's given, so teams point agents at a staging project when they want a hard safety boundary.

4. 4

   Operational checks on a schedule

   Scheduled reads pull project status, function lists and row counts so the team learns about drift or growth from a summary rather than from opening the dashboard.

The outcome

• Product and data questions are answered by an agent running a scoped query, not by a developer with a psql session.
• Supabase projects, functions and storage are read through the same tools as the rest of the stack, not a bespoke Management API client.
• AI agents reach the database through scoped, revocable OAuth access — never a raw connection string in a prompt.