Connectors

Admin → Connectors is where data sources are attached. Pick a template, fill in the fields, and the connector is created with a working configuration behind it.

PostgreSQL

MySQL

SQL Server

Snowflake

Databricks

Oracle

ClickHouse

Trino

Connecting one

1. Choose the template matching your database.
2. Fill in the required fields.
3. Choose Visibility.
4. Create the connector, then introspect it to populate the Registry.

Visibility

Every template offers a visibility setting:

• all — analysts can see and use this data source.
• admins — hidden from analysts, along with its tools.

Useful for a source holding sensitive data you want governable but not broadly answerable from, or for one you are still validating.

What happens under the hood

The database templates are not separate integrations. They are configurations of one read-only SQL server, app.mcp_db, pointed at your database with a DSN built from the fields you entered. The differences between templates are the DSN scheme and the field set, not the execution path.

That is why the same guarantee applies to all of them: only SELECT, WITH, EXPLAIN, SHOW, TABLE, and VALUES are allowed, and results are row-limited. See read-only by design.

Templates at a glance

Template	What it needs	Notes
PostgreSQL	host, port, database, user, password	The usual starting point
MySQL	host, port, database, user, password
ClickHouse	host, port, database, user, password
Microsoft SQL Server	host, port, database, user, password
Oracle	host, port, service name, user, password	Thin driver
Trino (PrestoSQL)	host, port, catalog, schema, user	Password optional
Apache Druid	host, port	Broker port, defaults to 8082
DuckDB	file path	Path is inside the backend container — mount a volume
Databricks SQL (JDBC)	host, HTTP path, token	For SQL warehouses
Databricks Genie	workspace URL, token	Remote HTTP MCP endpoint — a different path
Snowflake	account, user, password, database	Official MCP server; needs npx in the backend
Generic SQL	a SQLAlchemy DSN	Any dialect with a driver installed
Generic ODBC	host, port, database, user, password, driver	Needs pyodbc and the ODBC driver on the host

Full field detail is in the data source reference.

Requirements worth knowing before you connect

• DuckDB resolves its path inside the backend container, not on your machine. A local file will not be visible unless you mount it.
• Snowflake shells out to npx, which must exist in the backend container.
• Generic ODBC needs the driver installed on the host as well as pyodbc in the container.
• Databricks Genie and Snowflake use external MCP servers, so their permissions come from the credentials you supply — use a read-only role.

After connecting: introspect

Creating a connector does not register anything. Open Entity Registry → your connector and run Introspect to discover its tables and columns as entities.

Introspection is the step that makes the data usable, and it is the one most often skipped. See Entity Registry.

Something that is not a database

For anything that speaks MCP — a SaaS API, an internal service — see MCP servers.