Tool Catalog

AgentQ does its work by calling tools: named, structured operations that map to specific Qualytics capabilities. A tool is closer to an API call than to a free-form instruction. AgentQ cannot browse the internet or reach any service you have not configured, and it can only call the tools in this catalog. Some tools accept SQL for validation or asset creation. AgentQ does not receive source-query result sets; reporting tools return Qualytics metadata and permitted evidence.

The same catalog is available everywhere AgentQ is used:

The built-in AgentQ chat inside Qualytics.
External AI clients connected over MCP, such as Claude Desktop, ChatGPT, or Cursor.
The agentic endpoints used to build your own applications.

Two narrower assistants exist as well. The quality check and check template assistants that back the authoring forms use a small read-only subset of these tools and can only propose a definition for you to accept, never save one.

Whichever surface is used, every tool runs as the signed-in user and is subject to that user's team permissions. AgentQ has no identity or access of its own.

What Each Tool Shares with the Model

Because AgentQ sends tool results to a language model, it is worth knowing what those results contain. Every tool in this catalog falls into one of three categories.

Category	What reaches the model
No Sharing	No information about your data at all. Rule type catalogs, User Guide content, validation errors, and static guidance.
Metadata Shared	Information about your data, but no values from it. Datastore, table, and column names, data types, descriptions, tags, row and anomaly counts, quality scores, timestamps, and quality check definitions.
Source Data Shared	Everything in Metadata Shared, plus one or more individual values from your data, appearing either in a quality check failure explanation or as a profile statistic.

The categories are cumulative: each one includes everything above it in the table. They are also a control. A Manager or Admin chooses how far down this list AgentQ may go, so a tool below may be unavailable in your deployment even when your own permissions allow it. No Sharing is included automatically wherever an AI provider is configured. See AgentQ Access Controls.

Three things hold across the whole catalog, regardless of category:

No tool returns your records in bulk. There is no tool that hands the model source-query result sets or bulk sample records. Reporting rows contain Qualytics metadata, such as container names and anomaly counts. Values in the Source Data Shared category appear singly and incidentally, in the same way a steward sees a value when reading why a check failed.
Masked fields stay masked. AgentQ cannot reveal the underlying value of a field you have masked. See Masked Fields below.
Credentials are never shared. Connection details and secrets for your datastores and integrations are never sent to the model.

These categories describe what tools share from your connected data sources. Some tool results also identify platform users: user_profile returns your own profile, anomaly details include assignees' names and email addresses, comments carry their author's name, an anomaly's or container's change history names the person behind each change, and a check's details name its owner and default anomaly assignee. In the built-in AgentQ chat and through the agentic endpoints, your own name, email address, and role are also shared with the model at the start of the conversation, before any tool is called. The narrower authoring assistants do not do this, and over MCP nothing is shared until a tool runs.

The category shown for each tool below describes the most it can share. Most calls share less.

Semantic Reporting

Semantic reporting is the default in full AgentQ chat and MCP. Two tools handle schema discovery and reporting queries. They replace the earlier entity-specific discovery, detail, and history tools; those earlier names are no longer callable over MCP. Other read tools remain available for their richer reporting behavior.

The assistants embedded in check and template forms retain their bounded list_datastores, list_containers, and list_fields discovery tools, and the template assistant keeps list_quality_check_templates. Those names are form-only; the forms do not gain access to general reporting queries.

Tool	Data	Description
`describe_query_schema`	No Sharing	Discover the reporting resources, fields, filters, metrics, lifecycle rules, and limits available to you. Returns schema guidance rather than customer records.
`query_qualytics`	Source Data Shared	Read metadata, count and group authorized entities, or retrieve curated details, history, and daily insights. Each request is checked against the fields and evidence it uses. Metadata-only requests remain available at Metadata Shared.

The maximum classification of query_qualytics is Source Data Shared because profile statistics, comment text, anomaly descriptions, and historical evidence can reveal values. Selecting, filtering, grouping, or sorting by such data requires that level. A count that uses only metadata can run at Metadata Shared. These query-specific checks also apply when these two tools are called over MCP.

Semantic queries also expose current profiles, observability summaries, stored scores, derived counts, schedule cadence, next-run times and target selectors, operation outcomes and target containers, anomaly and partition comments with author metadata, quality checks with their field assignment and template linkage, templates with their lock state, detection-time failed-check evidence for anomalies, datastore-level weighted dimension scores, operation durations, and daily insight series for the platform, a datastore, a container, a field or a check. Recurring schedule listings, recent operations, comment threads, check listings, template listings, anomaly listings, quality scores and trend insights are read this way; there are no separate tools for them. Comment counts are metadata; comment text requires Source Data Shared. Field statistics require Source Data Shared and are null for masked fields, including when filtering or sorting.

See Semantic Reporting for query examples, lifecycle defaults, and limitations.

Discovery

Tool	Data	Description
`field_profile_describe`	Source Data Shared	Get a field's profile statistics. See Profile statistics.
`global_search`	Metadata Shared	Resolve datastore, container, and field names or IDs with parent context. Exact matches rank before prefixes and substrings.
`user_profile`	Metadata Shared	Get your own profile: your name, email address, role, favorites, and recent activity.

Quality Checks

Tool	Data	Description
`list_quality_check_specs`	No Sharing	List the available rule types and the properties each one accepts.
`create_quality_check`	Metadata Shared	Create a quality check.
`update_quality_check`	Metadata Shared	Update an existing quality check, including moving it between Draft and Active.
`validate_quality_check`	Metadata Shared	Test a proposed change to a check without saving it.
`dry_run_quality_check`	Metadata Shared	Run a check without recording anomalies and return counts only. It does not return the records that would fail.
`favorite_quality_check`	Metadata Shared	Add or remove a check from your favorites.
`delete_quality_check`	Metadata Shared	Archive a quality check, which can be restored later. By default this also removes the anomalies the check raised, unless you ask to keep them.
`create_quality_check_template`	Metadata Shared	Create a global template.
`update_quality_check_template`	Metadata Shared	Update a global template.
`delete_quality_check_template`	Metadata Shared	Archive a global template.

The authoring assistants add two propose-only tools, propose_quality_check and propose_quality_check_template, which return a suggested definition for your review and cannot save anything.

Computed Assets

Tool	Data	Description
`preview_query`	No Sharing	Check that a `SELECT` statement is valid before anything is created. See Validating SQL.
`create_computed_table`	Metadata Shared	Create a computed table in a JDBC datastore.
`create_computed_file`	Metadata Shared	Create a computed file in a file-based datastore such as S3, ADLS, or GCS.
`create_computed_join`	Metadata Shared	Create a join across containers, including across datastores.
`create_computed_field`	Metadata Shared	Add a derived or type-cast field to an existing container.
`update_computed_container`	Metadata Shared	Update a computed asset definition.
`delete_computed_container`	Metadata Shared	Delete a computed asset.
`update_computed_field`	Metadata Shared	Update a computed field definition.
`delete_computed_field`	Metadata Shared	Delete a computed field.
`validate_container`	Metadata Shared	Validate a computed asset definition without creating it.

Anomalies and Remediation

Tool	Data	Description
`update_anomaly`	Metadata Shared	Change an anomaly's status, for example to Acknowledged.
`bulk_update_anomalies`	Metadata Shared	Change the status of several anomalies at once.
`archive_anomaly`	Metadata Shared	Archive an anomaly with a final resolution status.
`create_comment`	Metadata Shared	Add a comment to an anomaly.
`update_comment`	Metadata Shared	Edit a comment.
`delete_comment`	Metadata Shared	Delete a comment.

Operations and Schedules

Tool	Data	Description
`run_sync`	Metadata Shared	Sync a datastore, discovering and registering its tables and views.
`run_profile`	Metadata Shared	Profile a container.
`run_scan`	Metadata Shared	Scan a container against its active quality checks.
`run_export`	Metadata Shared	Export anomalies, checks, or profiles from a datastore.
`run_materialize`	Metadata Shared	Materialize a computed asset.
`run_external_scan`	Source Data Shared	Check records you supply against a container's active checks. Unlike every other tool here, the records are provided to Qualytics by the caller rather than read from your datastore. Violations create anomalies, and the built-in chat receives their descriptions in the response.
`get_operation_status`	Metadata Shared	Poll an operation for progress and completion.
`abort_operation`	Metadata Shared	Stop a running operation.
`rerun_operation`	Metadata Shared	Run an operation again with the same settings.
`restart_operation`	Metadata Shared	Restart an operation.
`delete_operation`	Metadata Shared	Delete an operation record.
`create_schedule`	Metadata Shared	Create a recurring schedule for a sync, profile, scan, export, or materialize operation.
`update_schedule`	Metadata Shared	Change a schedule.
`run_schedule`	Metadata Shared	Trigger a scheduled operation immediately.
`set_schedule_deactivation`	Metadata Shared	Pause or resume a schedule.
`delete_schedule`	Metadata Shared	Delete a schedule.

Operations run in the background. These tools return an operation identifier straight away rather than waiting, so AgentQ polls for completion before moving on to a dependent step.

Stewardship

Tool	Data	Description
`update_field`	Metadata Shared	Update a field's description, tags, or status, including marking it masked.
`bulk_update_fields`	Metadata Shared	Update several fields at once.
`delete_field`	Metadata Shared	Exclude or delete a field. Fields are archived rather than removed by default.
`restore_field`	Metadata Shared	Return an excluded or masked field to active status.
`merge_fields`	Metadata Shared	Merge two fields in the same container.
`update_container`	Metadata Shared	Update a container's description, tags, or settings.
`bulk_update_containers`	Metadata Shared	Update several containers at once.
`favorite_container`	Metadata Shared	Add or remove a container from your favorites.
`delete_container`	Metadata Shared	Exclude or delete a container. A source table or file is archived, while a computed asset is permanently deleted.
`update_container_score_settings`	Metadata Shared	Change quality score weights and decay for a container.
`update_datastore`	Metadata Shared	Update a datastore's description, tags, or settings.
`favorite_datastore`	Metadata Shared	Add or remove a datastore from your favorites.
`assign_datastore_group`	Metadata Shared	Assign a datastore to a shared group, or remove it from one.
`update_datastore_score_settings`	Metadata Shared	Change quality score weights and decay for a datastore.

Insights and Scores

Stored quality scores, datastore-level weighted dimensions, daily insight series, and operation counts over time are read through query_qualytics. See Semantic Reporting. One assessment tool remains here.

Tool	Data	Description
`check_data_trust`	Metadata Shared	Assess whether one container can be trusted for AI or analytical use. Returns its quality score with the eight dimensions and their weights, active anomalies grouped by severity with a sample of the most severe, freshness readings, a description of the container and its fields, and its nearest lineage neighbors, together with guidance on how to weigh them. The verdict itself is written by the model from that evidence. See Trust Assessment.

Promotion

Tool	Data	Description
`promote_computed_table`	Metadata Shared	Copy computed table definitions to another datastore.
`promote_computed_file`	Metadata Shared	Copy computed file definitions to another datastore.
`promote_computed_field`	Metadata Shared	Copy computed field definitions to another container.
`promote_quality_check`	Metadata Shared	Copy quality checks to another container, landing as Draft by default.

Promotion copies definitions, not data.

Guided Workflows

Tool	Data	Description
`workflow_analyze_trends`	No Sharing	Guidance for analyzing quality score and anomaly trends over time.
`workflow_investigate_anomaly`	No Sharing	Guidance for investigating an anomaly and proposing remediation.
`workflow_interpret_quality_scores`	No Sharing	Guidance for reading quality scores in business terms.
`workflow_generate_quality_check`	No Sharing	Guidance for turning a business rule written in plain language into a check.
`workflow_transform_dataset`	No Sharing	Guidance for building a computed asset from a description.

These tools return a structured plan for approaching a task. They do not read your data or run other tools themselves. Any data involved arrives through the discovery, check, and anomaly tools that AgentQ calls as it follows the plan, and those calls are categorized above in the usual way.

Notifications and Tickets

Tool	Data	Description
`send_notification`	Metadata Shared	Send a message through a configured channel such as Slack, Microsoft Teams, email, webhook, or PagerDuty.
`create_ticket`	Metadata Shared	Create a ticket in Jira or ServiceNow, optionally linked to an anomaly.

Guidance

Tool	Data	Description
`search_userguide`	No Sharing	Search the Qualytics User Guide content bundled with your deployment. Results carry links to the matching guide pages hosted by your deployment. This is not a web search, and AgentQ has no other way to retrieve outside content.
`report_tool_constraint`	No Sharing	Report that a documented task cannot be completed reliably with the tools available, instead of guessing.
`report_no_direct_tool`	No Sharing	Disclose that no tool returns the requested answer directly, and describe the derivation AgentQ would use instead. See Answer Basis.

Masked Fields

Field masking is yours to control, and AgentQ respects it. Set a field's status to masked and the following applies:

Behavior	What happens
Distinct values in profile histograms	Replaced with a masked placeholder.
Values quoted in a quality check failure explanation	Replaced with `<masked>`.
Expected value checks	Never suggested for a masked field.
Unmasking	Not available to AgentQ in any form.

Exporting or materializing with masked values included is a separate, permission-gated action on the operation itself. A dry run (dry_run_quality_check) can also be asked to include masked values, and the request goes through the same permission checks; either way, the dry run only returns summary counts to the model, never the values themselves. AgentQ will not request masked values unless you ask in the conversation, and it will not carry that request forward into later turns.

Profile statistics

field_profile_describe returns the statistics Qualytics computes when profiling a field: minimum, maximum, mean, sum, quartiles, standard deviation, and length extremes. These scalar statistics are numeric, so for text fields only the length and count statistics are populated.

Histogram buckets are the one part of a profile that contains literal values from the field. They are omitted unless explicitly requested, and for a masked field they are replaced with a placeholder.

Validating SQL

preview_query confirms that a statement is valid before a computed asset is created. To do that it runs the statement against your datastore with a row limit, under your own permissions. The results are used only to determine whether the query works, and are then discarded. No rows and no column names are returned to the model, and nothing is created or saved.

Only SELECT statements and WITH clauses are accepted. See AgentQ Limits for the full list of blocked statements.

Permissions

Every tool checks your team permissions before it acts, using the same rules as the rest of the platform, and it runs as you rather than as a service account. For the permission each kind of tool needs, what happens when you lack it, and which deletions can be undone, see AgentQ Permissions.

Tool	Data	Description
`manage_tags`	Metadata Shared	Apply, remove, or replace tags on datastores, containers, fields, and checks.
`create_tag`	Metadata Shared	Create a global tag.
`update_tag`	Metadata Shared	Update a global tag.
`delete_tag`	Metadata Shared	Delete a global tag.