Collibra MCP tools

August 28, 2026

The Collibra MCP server exposes tools in three categories. The full list, including source code and schema definitions, is in the Git repository.

Note Most tools work with the default scopes granted to any authenticated Collibra user. Tools that require additional scopes are marked in the tables below. If a tool returns a permission error, you are missing the required scope. Contact your Collibra administrator to request access.

Read tools

Read tools return data from Collibra without making any changes.

Tool	Description	Extra scopes required	Uses Collibra Units?
`discover_business_glossary`	Ask questions about business terms and definitions using natural language.	`dgc.ai-copilot`	Yes
`discover_data_assets`	Query available data assets using natural language.	`dgc.ai-copilot`	Yes
`get_asset_details`	Retrieve detailed information about a specific asset by UUID. This includes the asset's assignable attribute schema, which lists all attributes the asset can hold, including empty ones.	N/A	No
`get_business_term_data`	Trace a business term back to its connected physical data assets.	N/A	No
`get_column_semantics`	Retrieve data attributes, measures, and business assets connected to a column.	N/A	No
`get_lineage_downstream`	Get downstream technical lineage (consumers) for a data entity.	N/A	No
`get_lineage_entity`	Get metadata about a specific entity in the technical lineage graph.	N/A	No
`get_lineage_transformation`	Get details and logic of a specific data transformation.	N/A	No
`get_lineage_upstream`	Get upstream technical lineage (sources) for a data entity.	N/A	No
`get_measure_data`	Trace a measure back to its underlying physical columns and tables.	N/A	No
`get_table_semantics`	Retrieve the semantic layer for a table: columns, data attributes, and connected measures.	N/A	No
`list_asset_types`	List available asset types.	N/A	No
`list_data_contract`	List data contracts with pagination.	N/A	No
`prepare_create_asset`	Enumerate available asset types and domains and hydrate the attribute and relation schema for a chosen pair. Use before `create_asset`.	N/A	No
`pull_data_contract_manifest`	Download the manifest for a data contract.	N/A	No
`search_asset_keyword`	Wildcard keyword search for assets.	N/A	No
`search_data_class`	Search for data classes with filters.	`dgc.data-classes-read`	No
`search_data_classification_match`	Search for associations between data classes and assets.	`dgc.classify`, `dgc.catalog`	No
`search_lineage_entities`	Search for entities in the technical lineage graph.	N/A	No
`search_lineage_transformations`	Search for transformations in the technical lineage graph.	N/A	No

Write tools

Write tools create, modify, or delete data in Collibra.

Tool	Description	Extra scopes required	Uses Collibra Units?
`add_data_classification_match`	Associate a data class with an asset.	`dgc.classify`, `dgc.catalog`	No
`create_asset`	Create a new asset of any type. Resolves asset type, domain, status, and attributes by name or UUID.	N/A	No
`edit_asset`	Edit an existing asset. Supports updating attributes, properties, relations, tags, and responsibilities. The following operations are supported: `update_attribute`, `add_attribute`, `remove_attribute`: Change append or clear an attribute value, such as `Definition` or `Note`. `update_property`: Rename the asset (`name`), change its `displayName`, or change its `statusId`. Status name and UUID are accepted. `add_relation`, `remove_relation`: Link or unlink the asset to another asset by relation role, such as `is synonym of`. `add_tag`: Append a free-text tag without replacing existing ones. `set_responsibility`: Assign a user or group to a resource role, such as `Steward` or `Owner`, by username, email, or UUID. `remove_responsibility`: Unassign a user or group from a resource role. This only applies to directly-assigned responsibilities, not inherited ones.	N/A	No
`init_data_contract`	Initialize a new data contract asset governing a Data Product Port, with an optional initial manifest.	`dgc.data-contract`	No
`push_data_contract_manifest`	Upload a manifest for a data contract.	`dgc.data-contract`	No
`remove_data_classification_match`	Remove an association between a data class and an asset.	`dgc.classify`, `dgc.catalog`, `dgc.data-classes-edit`	No

Experimental tools

Experimental tools are opt-in and off by default. They may change or be removed without a deprecation cycle. To enable them, pass --experimental=<name> at startup, set the COLLIBRA_MCP_EXPERIMENTAL environment variable, or add the feature name under mcp.experimental in your mcp.yaml file. For configuration details, see the Git repository.

Note Experimental tools apply to the local MCP server only. The remote server’s tool set is managed by Collibra.

Feature flag	Tools added	Description
`data-quality`	`create_data_quality_job`	Sets up automated DQ monitoring on a database table and queues the first run. This single tool spans discovery, preview, and creation.
	`create_data_quality_rule`	Creates a single rule on an existing job. `monitorType`: Choose `FREEFORM_SQL` for a full SQL query or `SIMPLE_SQL` for a single-column check. The rule is active and not suppressed by default. Confirm checkpoint: Set `confirm=false` to preview the rule and SQL. Set `confirm=true` to create the rule.
	`deploy_data_quality_rule_template`	Instantiates a rule template as concrete rules across one or more job/column targets in bulk. The DQ engine resolves dialect-specific SQL and names each rule `{templateName}_{columnName}`. Confirm checkpoint: Set `confirm=false` to preview the template and targets. Set `confirm=true` to deploy.
	`dq_cancel_job_run`	Cancels an in-progress run. You can specify either the run's `jobRunId` or `jobName`. If you provide a `jobName`, the tool identifies cancellable runs. If there is only one eligible run, it is canceled directly. Otherwise, the tool returns a list of candidates for you to choose from. Canceling a run is irreversible. There is no confirm checkpoint. The terminal or non-terminal state check serves as the safety mechanism.
	`dq_delete_job`	Permanently deletes a job definition with all of its runs, rules, monitors, and results. This cannot be undone. If a run is in progress, the service may refuse the delete, so you must cancel it first with `dq_cancel_job_run`. Confirm checkpoint: When `confirm=false` (default), the operation is read-only. It retrieves the job details and returns a summary, including the job type, Edge site, connection, schema or table, source query, and schedule. When `confirm=true`, the job is deleted.
	`dq_delete_job_run`	Permanently deletes a completed run along with its per-run results (profile, scan, monitor, rule, alert output). You can specify either the run's `jobRunId` or `jobName`. If you provide a `jobName`, the tool identifies eligible terminal runs but does not delete them directly. You must re-run the operation with the selected `jobRunId` to proceed. Only terminal runs (FINISHED, CANCELLED, or FAILED) can be deleted. To delete an in-progress run, cancel it first using `dq_cancel_job_run`. Confirm checkpoint: When `confirm=false` (default), the operation is read-only and returns the run's details. When `confirm=true` (used with `jobRunId`), the run is deleted.
	`dq_get_job`	Reads the full definition of a single data quality job by name. The definition includes the type (PUSHDOWN or PULLUP), Edge site, connection, schema/table, source SQL, run-date window, configured monitors (adaptive and custom), notifications flag, and schedule. An exact name match is tried first. If none is found, jobs whose name contains the text are offered as candidates to disambiguate. Read-only with no confirm checkpoints.
	`dq_get_job_run`	Retrieves the full details of a single job run using its `jobRunId`. The details include lifecycle status or activity, timing, and, once the job reaches a terminal state (FINISHED, CANCELLED, or FAILED), the score, row count, execution time, and a per-monitor breakdown (adaptive and custom) contributing to the score. Read-only with no confirm checkpoints.
	`dq_update_job`	Partially updates an existing job identified by its `jobName`. Provide only the fields you want to change; all other fields remain unchanged. Use this instead of deleting and recreating a job, which would erase the run history, results, and learned monitor baselines. You can update the following fields: scan SQL, run-date window, recurring schedule, monitor set and adaptive baseline, notifications, PUSHDOWN compute settings, PULLUP sizing, and data location (for a moved or renamed table). Merge behavior: Most settings are updated field by field. However, monitors, the schedule, and notifications are replaced entirely when modified. Re-supply any parts you want to keep. Limitations: The job type (PUSHDOWN or PULLUP) cannot be changed. Data quality rules must be managed using `create_data_quality_rule` or `deploy_data_quality_rule_template`. Confirm checkpoint: By default, `confirm=false` returns a before-and-after comparison and the exact request without making changes. Set `confirm=true` to apply the changes.
	`find_data_quality_rules`	Searches existing rules across jobs. Filter by exact `jobName`, `columnName`, or a rule-name substring. Combine `jobName` and `columnName` to check whether a rule already exists on a target column before creating a new one. Returns each rule's job, column, type, status, and SQL. Paginated (`offset`/`limit`).
	`generate_data_quality_rule_sql`	Uses Collibra DQ AI (Text2SQL) to convert a plain-language description of a check into rule SQL. This enables you to create rules without manually writing SQL. Always validate the generated SQL using `validate_data_quality_rule` before applying it. Requires `edgeSiteId` and `connectionId`, both available from `create_data_quality_job`.
	`get_data_quality_rule`	Reads the full definition of a single rule on a job, including its type (`FREEFORM_SQL` or `SIMPLE_SQL`), SQL, filter clause, tolerance (number of failing records allowed before the rule fails), and active/suppressed state.
	`get_data_quality_rule_results`	Reads a rule's per-run history, including a score (0–100), breaking (failing) and passing record counts, pass/fail status, and any exception. Returns newest first by default. Paginated (`offset`/`limit`).
	`get_data_quality_rule_template`	Reads a single template by name. The template includes parameterized SQL, dialect, dimensions, default tolerance, whether it's built-in (out-of-the-box), and how many rules have been deployed from it.
	`list_data_quality_rule_templates`	Lists all available rule templates, including built-in (out-of-the-box) and custom. Each template is a parameterized SQL pattern deployable via `deploy_data_quality_rule_template`. Optional filters: `name`, `dimension`, `isSystem` (built-in or custom).
	`validate_data_quality_rule`	Validates a rule's SQL against the source database before saving or running it, catching malformed rules up front. Returns a pass/fail verdict plus the engine's validation message. Requires `edgeSiteId`, `connectionId`, and `schemaName`, all available from `create_data_quality_job`.
`skills`	`list_collibra_skills`, `load_collibra_skill`	Serves an embedded catalog of Markdown guides documenting multi-step Collibra workflows — discovery, lineage, asset creation, and more — for the connecting AI model.