Data source templates
Reference for the templates under Admin → Connectors.
How they work
Section titled “How they work”The database templates are not separate integrations. Each is a configuration of
one read-only SQL server, app.mcp_db, pointed at your database with a DSN built
from the fields you enter. The difference between templates is the DSN scheme and
the field set, not the execution path.
Templates
Section titled “Templates”PostgreSQL
Section titled “PostgreSQL”Host, port (default 5432), database, user, password, and an optional SSL mode.
Host, port (default 3306), database, user, password.
ClickHouse
Section titled “ClickHouse”Host, port (default 8123), database, user, password.
Microsoft SQL Server
Section titled “Microsoft SQL Server”Host, port (default 1433), database, user, password.
Oracle
Section titled “Oracle”Host, port (default 1521), service name, user, password. Uses the thin driver — no Oracle Instant Client required.
Trino (PrestoSQL)
Section titled “Trino (PrestoSQL)”Host, port (default 8080), catalog, schema, user. Password optional.
Apache Druid
Section titled “Apache Druid”Host and broker port (default 8082). Note this is the broker port, not the router port.
DuckDB
Section titled “DuckDB”A file path. Resolved inside the backend container, not on your machine — a local file is not visible unless you mount it as a volume.
Databricks SQL (JDBC)
Section titled “Databricks SQL (JDBC)”Host, HTTP path, and a token. SQL warehouse access.
Databricks Genie
Section titled “Databricks Genie”Workspace URL and token. A remote HTTP MCP endpoint, not the JDBC path — its permissions are whatever the endpoint enforces.
Snowflake
Section titled “Snowflake”Account, user, password, database. Runs the official Snowflake MCP server,
so the backend container must have npx available.
Generic SQL
Section titled “Generic SQL”A SQLAlchemy DSN. Any dialect whose driver is installed in the container. This is the route for BigQuery and Redshift, which have no dedicated template.
Generic ODBC
Section titled “Generic ODBC”Host, port, database, user, password, and an ODBC driver name. Requires
pyodbc and the corresponding driver installed on the host as well as in the
container.
Requirements that catch people out
Section titled “Requirements that catch people out”| Template | Requirement |
|---|---|
| DuckDB | Path is inside the backend container — mount a volume |
| Snowflake | npx must exist in the backend container |
| Generic ODBC | ODBC driver on the host and pyodbc in the container |
| Druid | Use the broker port (8082), not the router port |
| Databricks Genie, Snowflake | Permissions come from the credentials — use a read-only role |
BigQuery and Redshift
Section titled “BigQuery and Redshift”Neither has a dedicated template. Use Generic SQL with a SQLAlchemy DSN, or Generic ODBC if that is what your environment already has drivers for.
Read-only enforcement
Section titled “Read-only enforcement”All in-product SQL templates share the same enforcement: only SELECT, WITH,
EXPLAIN, SHOW, TABLE, and VALUES are permitted, results are row-limited,
and identifiers are quoted. The MCP-based templates (Databricks Genie, Snowflake)
and anything attached as an MCP server are outside that path —
their permissions are whatever the credentials or endpoint allow.
Full detail in read-only enforcement.