> ## Documentation Index
> Fetch the complete documentation index at: https://docs.genesiscomputing.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Collibra Tools

> Search and inspect Collibra catalog metadata through a saved, read-only connection.

## `collibra_tools`

Use the `collibra_tools` group to read Collibra assets and related catalog
metadata. It is read-only: it never creates or changes Collibra records. Enable
the tool group for the agent and configure a named Collibra connection the
requesting user and agent may use. See [Collibra integration](https://docs.genesiscomputing.com/docs/Integrations/Data%20Sources/Collibra/Collibra)
and [Collibra connections](https://docs.genesiscomputing.com/docs/Setup/Connections.md#collibra-connections).

### `collibra_tool`

| Parameter | Type | Default | Description |
| - | - | - | - |
| `action` | string | required | Ordinary REST actions: `TEST_CONNECTION`, `SEARCH_ASSETS`, `LIST_COMMUNITIES`, `LIST_DOMAINS`, `GET_RELATIONS`, `GET_ATTRIBUTES`, `LIST_TYPES`, `GET_ASSET`, or `REQUEST`. Ingestion actions: `FIND_TABLE`, `GET_TABLE_METADATA`, `DISCOVER_TABLES`, or `BULK_TABLES`. |
| `connection` | string | required | Named Collibra connection, such as `collibra.catalog`. |
| `asset_id` | string or null | `null` | `GET_ASSET`: asset UUID. `GET_TABLE_METADATA`: explicitly selected table UUID. |
| `path` | string or null | `null` | `REQUEST` only: absolute REST v2 path under `/rest/2.0/`, without a host or query string. Supply filters separately in `params`. |
| `method` | string | `GET` | Only `GET` is accepted. |
| `params` | object or null | `null` | REST filters for ordinary actions; ingestion inputs are `table_name`, `table_names` or `table_ids`, `domain` or `domain_id`, and `source_system`. Do not include credentials or request headers. |
| `offset` | integer | `0` | Ordinary REST actions only: non-negative offset for paged reads. |
| `limit` | integer | `100` | Ordinary REST actions only: page size, capped at `500`; ingestion paging is automatic. |
| `bot_id`, `user_id` | implicit | current context | Injected by Genesis and used for connection access checks. |

Ingestion fields inside `params`:

| Field | Type | Default | Use |
| - | - | - | - |
| `table_name` | string | none | Required by `FIND_TABLE`; exact table name to look up. |
| `domain` | string | none | Optional exact domain name for `FIND_TABLE` and `BULK_TABLES`; required as one of the scope selectors for `DISCOVER_TABLES`. |
| `domain_id` | UUID string | none | Optional domain scope; required as one of the scope selectors for `DISCOVER_TABLES`. |
| `source_system` | string | none | Optional exact source-system name; required as one of the scope selectors for `DISCOVER_TABLES`. |
| `table_names` | string array | none | For `BULK_TABLES`, request candidate lookups by name; provide 1–500 entries and omit `table_ids`. |
| `table_ids` | UUID string array | none | For `BULK_TABLES`, request full metadata for explicitly selected tables; provide 1–500 entries and omit `table_names`. |

Named list and search actions return the original REST response, including
`results` and paging information. Use the returned paging values to request
subsequent pages explicitly. `GET_ASSET` reads one asset by UUID. `TEST_CONNECTION`
authenticates and reports the number of assets visible to the connection; zero
visible assets is a failed visibility check.

For ingestion, `FIND_TABLE` accepts `params.table_name` and optionally a `domain`,
`domain_id`, or `source_system`; it returns matching candidates and marks ambiguous
results as requiring selection. Pass the chosen candidate's UUID as `asset_id` to
`GET_TABLE_METADATA`. `DISCOVER_TABLES` requires a named `domain`, `domain_id`, or
`source_system` and returns candidate summaries. Lookup and discovery refuse scopes
over 2,000 typed tables. `BULK_TABLES` requires exactly one of `table_names` or
`table_ids`, with 1–500 entries. Names return candidate lookups; explicit UUIDs return
full metadata. Check `pipeline_ready` and `blocked_reasons` before building a pipeline;
unconfirmed approval, column order, missing metadata, or an unconfirmed table path is
a blocker. Configure nonstandard Collibra selectors through the connection's optional
`ingestion_model` field; see [Collibra ingestion](https://docs.genesiscomputing.com/docs/Integrations/Data%20Sources/Collibra/Collibra#collibra-ingestion).

```python theme={null}
collibra_tool(action="FIND_TABLE", connection="collibra.catalog",
              params={"table_name": "Customer", "domain": "Analytics"})
collibra_tool(action="GET_TABLE_METADATA", connection="collibra.catalog",
              asset_id="<selected-table-uuid>")
```

```python theme={null}
collibra_tool(
    action="SEARCH_ASSETS",
    connection="collibra.catalog",
    params={"name": "Customer"},
    offset=0,
    limit=100,
)
```

Ordinary calls normally make one Collibra GET and make a token POST only when the
shared OAuth token has expired. A request may add up to two transient retries
and one refresh/retry after HTTP 401, with bounded waits. Access is resolved for
each call; cached tokens do not bypass connection authorization. `REQUEST` is
restricted to GET paths under `/rest/2.0/` and does not accept a URL, body, or
authorization header. Ingestion reads pages sequentially (up to 100 pages of 500
per collection), then may make hierarchy requests and asset/attribute reads per
column; it can require many REST calls and makes no writes. See [Collibra
integration](https://docs.genesiscomputing.com/docs/Integrations/Data%20Sources/Collibra/Collibra) for the read-only
boundary and access model.
