Next.js + Prisma

EnvironmentCopy Anchor Link

DATABASE_URL="postgres://user:password@host:6432/db?sslmode=verify-full"
DIRECT_URL="postgres://user:password@host:5432/db?sslmode=verify-full"

capydb link writes exactly this when it detects Prisma - DATABASE_URL points at the pooled endpoint and DIRECT_URL at the direct one. If you deploy on Vercel with the Vercel integration, the pushed names are DATABASE_URL (pooled) and DATABASE_URL_UNPOOLED (direct); map directUrl accordingly.

Prisma schemaCopy Anchor Link

This page shows the Prisma 6 datasource. For Prisma ORM 7, where the connection URL moved out of the schema into prisma.config.ts and the client takes a driver adapter, run capydb init prisma in the linked project: it writes prisma.config.ts (migrations on the direct URL), prisma/schema.prisma, and a client on @prisma/adapter-pg over the pooled URL - see the CLI reference.

datasource db {
  provider  = "postgresql"
  url       = env("DATABASE_URL")
  directUrl = env("DIRECT_URL")
}

url is what Prisma Client uses at runtime - pooled, safe for serverless. directUrl is what prisma migrate and prisma db push use - direct, because migrations need session-level features the transaction pooler does not provide.

Run itCopy Anchor Link

npx prisma migrate dev    # uses DIRECT_URL
npx prisma studio         # uses DIRECT_URL
next dev                  # app traffic over DATABASE_URL

PitfallsCopy Anchor Link

• Migrations against the pooled URL fail in confusing ways (advisory lock errors, prepared statement errors). Always set directUrl.
• Serverless connection storms: Prisma Client per-invocation on the direct URL exhausts the plan's connection budget fast. The pooled URL exists for exactly this.
• Preview deployments: a preview database has its own pair of URLs. The Vercel integration scopes them to the git branch automatically; with the GitHub Action they are exported into the workflow env.