Vanta

Create a new data processing activity.

Create one or more custom controls in the organization's compliance program. Validates each control first and creates only the valid ones. `owner` and `customFields` take ids, never names, and you must resolve them yourself before calling this tool — never ask the user for an id. Resolve an owner with `searchUserAccounts` for a person or `searchUserAccountTeams` for a team; personnel records are not user accounts and cannot own a control, so never resolve an owner with `listPersonnel` or `listPersonnelGroups`. If the user names a field that does not correspond to one of this tool's own inputs, treat it as a custom field: look it up with `listControlCustomFields` to get its `customFieldId` and, for a picklist, the id of the option they named, then pass both in `customFields` on this call. This tool can PARTIALLY succeed: each requested control is either created (returned in `controls`) or rejected (returned in `errors`, grouped by the failed control's `shorthandName`, each holding a nested list of problems — every problem has a machine `type` and a human-readable per-problem `message`; when that problem's `values` is empty, infer the offending input from its own `message`). `createdCount` (= `controls.length`) is the number actually created, `failedCount` (= the number of `errors` entries) the number rejected, and `requestedCount` the number of inputs. `createdCount + failedCount` can be LESS than `requestedCount`: inputs that share a `shorthandName` collapse into a single `errors` entry, so a `shorthandName` repeated across N inputs means N inputs collided and only one of them can keep that ID. After every call you MUST report the outcome to the user from `createdCount` and `errors` — never assume all requested controls were created, and never continue without reporting it. The top-level `message` is a prewritten one-line summary (e.g. "Created 2 of 4 controls; 2 had errors.") you may show as-is, but it carries no per-control detail — take specifics from `createdCount`/`errors`, not from it. Never resubmit a control already in `controls`; resubmit a rejected one only after changing the offending value. You MUST resolve every `errors` entry before moving on. Resolving means surfacing the failure and getting the offending value changed — not resubmitting: a validation error recurs unchanged until that value changes, so propose the change to the user and wait for their go-ahead rather than retrying on your own initiative. Error examples: ID collision (duplicated among the inputs, or already in use by an existing custom or Vanta built-in control), invalid framework or section, unknown owner, etc. Fix an ID collision by proposing a new, meaningful `shorthandName` to the user — never by appending a disambiguating suffix (e.g. "(1)"/"(2)" or "A"/"B"). `listControls` surfaces IDs already in the program, but a submitted ID can still collide with a built-in Vanta control not yet in the program, so a `shorthandName` collision can come back even after listing. Surface anything you cannot resolve to the user.

Create a new data controller.

Create a new data subject request (DSR).

Create a new assessment in the organization's privacy program. The only supported assessment types are data protection impact assessment (DPIA), legitimate interest assessment (LIA), and transfer risk assessment (TRA). The user must specify the name of the assessment and the type in their request.

Create a new in-platform draft version of a privacy assessment. Omit templateId to carry the current in-platform answers forward, keeping the template the same (in-platform assessments only). Provide a templateId to start from a blank compatible template with no prior answers — required for uploaded assessments, which have no in-platform questionnaire to carry forward.

Create a new processing category.

Create a new risk register with a given name and optional description. Names must be unique within the domain (case-insensitive).

Create one or more tag categories for this organization. Each category groups related tags. Category display names must be unique.

Deactivate one or more controls in the organization's compliance program.

Use this tool when the user asks you to deactivate a document in Vanta. Deactivation stops the document from appearing as due and stops its renewal schedule. Call only when you know the document's evidenceRequestId and the user has supplied a reason. Check the document's details first. If it is already deactivated, tell the user instead of calling this tool. Deactivation always needs a reason. Ask the user for one and pass it through verbatim — never invent one, and never reuse a reason from an earlier call. By default a document stays deactivated indefinitely. Only set reactivateOn when the user names a date or a period to pause for; never guess one. The date must be after today. If the requested date is today or in the past, ask for a future date instead of calling the tool. Talk to the user only in the product's vocabulary: Document, Deactivate, Reason.

Deactivate monitoring for one or more automated tests. Use when the user wants a test to stop running and alerting — e.g. a known false positive, a not-applicable control, or temporarily silencing a noisy or failing test. This tool only supports automated tests (items with type "TEST"). Do not call it for evidence requests or documents (items with type "DOCUMENT"). Provide the test IDs and a short reason (recorded on the change). Deactivation is permanent by default; to deactivate temporarily, set expirationDays to the number of days after which monitoring automatically resumes. If the user hasn't made clear whether the deactivation should be temporary or permanent, ask before proposing.

Permanently delete one or more APPROVED answer library entries. Provide the entryIds to delete; deletion cannot be undone. IMPORTANT: this operates on APPROVED answer library entries, NOT pending ones — the entryIds must come from getAnswerLibraryQuestionAnswers, not getPendingAnswerLibraryQuestionAnswers. Only call after the user explicitly asks to delete approved entries.

Remove common control to framework control mappings in the organization's compliance program.

Delete a single processing activity.

Delete a single data controller.

Permanently deletes an entire policy — every version, every locale, and its associated program state. This cannot be undone. `policyType` must come from a prior policy lookup, never guessed.

Delete a privacy assessment and ALL of its versions — every version, including approved historical ones, not just the latest. This is destructive and irreversible.

Delete a single processing category.

Delete a single risk register by its ID. Always look up the register first so you can pass the current ID. If the user's request matches more than one register, ask the user to disambiguate before calling this tool — do not guess. Cannot delete the Enterprise risk register or the last register of its type.

Delete one or more tag categories and all tags within them. This also removes all tag attachments for those tags. This action cannot be undone.

Returns a pre-signed URL for a policy document

Edit a single APPROVED answer library entry. Provide the entryId plus only the fields you want to change (question, answer, expiration, and/or owner); any field you omit keeps its current value. Pass expiresAt or ownerUserId as null to clear them. Returns the updated entry. IMPORTANT: this operates on an APPROVED answer library entry, NOT a pending one — the entryId must come from getAnswerLibraryQuestionAnswers, not getPendingAnswerLibraryQuestionAnswers. Tags are managed with the separate add/removeAnswerLibraryTags tools. Only call after the user explicitly asks to change an approved entry.

Edit an audit's observation dates or auditor early-access date. Send only the dates the customer wants changed. Changing observation dates moves the audit's evidence window. Does not change audit names, auditors, report dates or completion dates. Report the returned persisted dates, which may be normalized, rather than the requested ones.

Get detailed information about a specific access review. In account snapshot mode, the response includes: - accountAccessRecommendation on each snapshot (APPROVE / DENY / NEEDS_REVIEW / NO_INSIGHT with reasons). null if not yet computed. Recommendations are advisory and do not replace human judgment. - aiRecommendationJobStatus on the top-level response: PROCESSING, QUEUED, FAILED, COMPLETED, or null. **Example queries:** - "Which systems completed review in access review X?" -> accessReviewId: X, vendorStates: ["COMPLETED"] - "Which systems are still pending?" -> accessReviewId: X, vendorStates: ["NOT_STARTED", "STARTED"] - "Show flagged accounts in system Y of review X" -> accessReviewId: X, vendorId: Y, accountSnapshotFlags: ["OWNER_TERMINATED", "OWNER_INACTIVE"] - "What's the recommendation for John's Salesforce access?" -> accessReviewId: X, vendorId: Y

Retrieve answer library entries. Provide entryIds to fetch specific entries directly by their IDs — in that case all other filter and sort arguments are ignored. Omit entryIds to list entries with search and filters. Returns each entry's question, answer text, formatted answer, source, owner, and timestamps.

Get the tags currently attached to one or more answer library entries. Returns a map of entry ID to its tag identifiers (category + id pairs). Use this to inspect which tags are applied before adding or removing tags.

Get comprehensive details for one or more automated tests by providing test ID(s). If you need details for multiple test IDs pass them all in a single call. Returns, per test, test metadata, outcome, deactivation status (isDeactivated), the due date for a failing test (soonestRemediateByDate), related controls, failing entities (up to 20 by default, or all if includeAllFailingEntities is set), optionally owner information, and the configured parameters for parameterized tests (e.g. a threshold like the access key rotation period), or null when the test has none.

Get or generate AI control-to-policy mapping suggestions for one or more policies. Non-blocking and idempotent: it starts generation for policies that have not run yet and returns the current rollup. status is in_progress while any policy is still generating; once all policies are terminal it is success (all completed), partial_success (some completed, some failed), or failure (none completed). Completed policies include their hydrated suggestions inline in suggestionsByPolicyType, each with a suggestionId that createPolicyControlMappings accepts. Failed policies are terminal (the underlying job retries internally) and are listed in failures. Generation typically takes a few minutes: when status is in_progress, report back to the user and check again in a few minutes rather than calling this repeatedly in a tight loop. For DOCX-backed policies, pass the same uploadedDocumentId used to create the policy; for PDF-backed policies the policy type is enough.

Get full details of one or more data processing activities, including current custom field values, categories of personal data, vendors, joint controllers, data locations, and security measures. Look up activities by ID or directly by name. Name matching is a case-insensitive substring search over activity names and processing purposes, so a single name may return several activities.

Retrieves the full details of a single data subject request (DSR).

List the domain's enabled data subject relationship options. Call this before creating or updating a data subject request (DSR) to get the allowed values for its relationship. Each option has two fields: pass `dataSubject` as the DSR's relationship value on create/update; `displayName` is only the human-facing label to show the user when asking them to pick one.

Get comprehensive details and status information for evidence requests in Vanta.

Retrieve the requirements for one or more frameworks. Each entry is one read: a framework id, a scope, and optional catalog variants. The same framework may appear more than once. MY_PROGRAM is this domain's current security posture on the framework. CATALOG is the framework's published requirements, independent of this domain. For CATALOG, variants selects a published tier; omit it for that framework's defaults. Some frameworks accept only one catalog variant per entry; compare catalogs by sending one entry per variant. Framework ids are canonical (e.g. 'pciDss4', 'iso27001', 'nist53').

Get all sections for a specific framework

Get knowledge base insights including coverage analysis and answer library conflicts. This is useful for: - "How well does our knowledge base cover security questions?" - "What security areas need more documentation?" - "Are there any conflicting answers in our knowledge base?" - "Show me knowledge base health"

Get the tags currently attached to one or more knowledge base resources. Returns a map of resource ID to its tag identifiers (category + id pairs). Use this to inspect which tags are applied before adding or removing tags.

Retrieve pending answer library entries. Provide a non-empty pendingAnswerIds to fetch specific pending entries directly by their IDs — in that case all other filter arguments are ignored. Omit pendingAnswerIds to list entries with search and filters. Returns each entry's question, answer text, formatted answer, source, tags, and timestamps.

Get the list of groups that a specific person belongs to. Returns up to 50 groups per request. Use pagination to fetch more. Returns person's groups with: - id: unique group identifier - name: group name - createdAt: date and time the group was created, or null for legacy groups - description: optional group description - personnelCount: number of people in the group - source: name of the service that created the group (e.g., IDP names like "Google Workspace", "Okta", or "Vanta" for manually created groups) - lastRefreshedFromIdp: ISO timestamp when the linked IdP group was last refreshed, or null for manually created groups - pointOfContactName: name of the group's point of contact (if assigned) - pointOfContactEmail: email of the group's point of contact (if assigned) Also returns pagination info: - groupsTotalCount: total number of groups matching the filter - pageInfo.hasNextPage: whether more groups exist - pageInfo.endCursor: pass this as "after" parameter to fetch the next page **Example queries:** - "Show me the groups for personnel record user-abc-123" -> Use personnelRecordId: "user-abc-123" - "What groups does the person with id rec-456 belong to?" -> Use personnelRecordId: "rec-456" - "Next page of groups" -> Use personnelRecordId and after: pageInfo.endCursor from previous response

Get aggregate acceptance statistics for one or more security policies. When to use this tool: - Aggregate personnel policy acceptance questions like counts and rates - NOT for specific personnel: "Did [name] accept the policy?" - NOT for personnel task completion: "Did [name] complete their onboarding task?" Tool output (per policy) will contain an acceptanceStats object with the following: - numUsersAccepted: Count who accepted - numUsersRequired: Count who need to accept - acceptanceRate: percentage of required users who have accepted (e.g., "85%") - isFullyAccepted: true if all required users have accepted