Most database servers assume the database already exists. This one creates it, mints credentials, creates tenants and then runs SQL — which suits Nile's tenant-per-customer model, where spinning up isolated data is routine rather than a one-off. Two things to hold onto: a new credential's password is shown once, and `delete-database` is as reachable as `list-databases`.
The Nile platform's MCP server. It covers the whole lifecycle of a Nile Postgres database: create it, get credentials, inspect schema, run queries, and manage the tenants that Nile's multi-tenant model is built around.
- Databases — `create-database` (region `AWS_US_WEST_2` for Oregon or `AWS_EU_CENTRAL_1` for Frankfurt), `list-databases`, `get-database`, `delete-database`
- Credentials and regions — `create-credential`, which returns a one-time password you must save when shown, `list-credentials`, `list-regions`
- SQL — `execute-sql` against a named database, with credentials created automatically when you do not pass a connection string; results come back as a markdown table with row counts and errors carry hints
- Schema — `read-resource` for a table's columns, types, primary keys, indexes and foreign keys; `list-resources` for every table and view
- Tenants — `list-tenants`, `create-tenant`, `delete-tenant`
A Nile API key and workspace slug, set as `NILE_API_KEY` and `NILE_WORKSPACE_SLUG` — create the key from your workspace's Security settings. Install `@niledatabase/nile-mcp-server`, build with `npm run build`, then register `node dist/index.js` in Claude Desktop or Cursor with those two variables in the env block. Stdio is the default; set `MCP_SERVER_MODE=sse` to serve over HTTP instead, with an events endpoint at `/sse` and commands posted to `/messages`. MIT.
One command plus a key — npm install @niledatabase/nile-mcp-server, then supply credentials