Data sources

Two read-only operations work with your organization's data sources. data.sources.browse lists what is in the database an installed query data source reads: its schemas, a schema's tables, or a table's columns. It returns names and types, never a row. data.sources.connectorProfiles checks data source manifests: it compiles the manifests you send and lists what is wrong with each one that does not compile.

import { createHttpPlatformTransport, createScrydonSDK } from '@scrydon/sdk/platform'

const platform = createScrydonSDK(
  createHttpPlatformTransport({
    baseUrl: 'https://scrydon.com',
    headers: () => ({ authorization: `Bearer ${accessToken}` }),
  })
)

const schemas = await platform.data.sources.browse({
  sourceId: 'orders-db',
  connectionRef: connectionId,
  browse: { level: 'schemas' },
})

const { profiles, refusals } = await platform.data.sources.connectorProfiles({
  sources: [{ sourceId: 'draft', manifestSnapshot: manifest }],
})

Over HTTP, both need the tables:read scope. They are read-only and safe to retry.

In a workflow, select the Scrydon tools Browse data source database and Check data source manifests in an Agent block's tools. Inside a Scrydon tool, the same operations are ctx.data.sources.browse and ctx.data.sources.connectorProfiles. Cortex runs both tools without asking for approval, because they only read. On that path no per-operation scope is checked: what a browse may reach is set by the run's grant, by the person's use of the connection (checked when the run acts for a person), and by the data source being set up to read through that connection.

Browse

browse.level picks what is listed:

• Names come in pages. limit is 1 to 500 and defaults to 100. To read the next page, pass the previous page's nextCursor as cursor. A nextCursor of null means there is no next page.
• A schema or table name is lowercase. It is a letter or _, then up to 62 letters, digits or _: the names a data source manifest can select.
• rowCountEstimate is the database engine's statistic, not a count. It is null where the engine does not expose one.
• Personal data is judged from names and types. No row is read to decide it.

The read runs as the person who calls it, in the workspace environment the call is made in. The person must be allowed to use the connection. The data source must be installed in that workspace environment and set up to read through that connection.

Check manifests

The result is { profiles, refusals }.

• profiles has one entry for each distinct connector profile. Its sourceIds lists every manifest that compiled to it. profile carries the profile's version and digests, and lineage lists where each column's value is read from.
• refusals has one entry for each manifest that did not compile, with its sourceId and a reason: invalid_manifest, incoherent_profile or unpinned_format. issues lists up to 100 problems, each with the field path and a message. issuesTruncated is true when there were more.

The operation reads nothing of your organization. It compiles only the manifests you send.

@scrydon/sdk/platform exports the bounds (DATA_SOURCE_BROWSE_LIMITS, CONNECTOR_PROFILES_LIST_LIMITS) and the levels (DATA_SOURCE_BROWSE_LEVEL). Unknown fields, including an organization or workspace field, are refused.

Errors