MCP App Store
by KeenEthics
Drata icon

Drata MCP Server for Claude and ChatGPT

By Drata Inc.

Overview

Drata's MCP server allows automating all aspects of GRC covering the full lifecycle of controls, policies, risks, evidence, frameworks, vendors, personnel, and devices directly inside the Drata platform.

How to connect

  1. Claude: Open the Drata listing in the Claude connectors directory and choose Connect. Claude listing Drata exposes 36 tools here.
  2. ChatGPT: Open the Drata listing in the ChatGPT app directory and choose Connect. ChatGPT listing Drata exposes 49 tools here.

Tools

Drata_createBackgroundCheck

ChatGPT
File a Background Check for a personnel record, evidenced by a URL. WRITE operation. Requires the create:background-checks OAuth scope AND the Create ability on Personnel. Both gates run; the scope alone is not sufficient. `url` is the only evidence source. Drata records it as-is and never fetches or scans it. This tool does not upload files; if the user has a file rather than a link, tell them it must be hosted somewhere Drata can reference, and do not invent a URL. ONE check per call. There is no multi-URL form. To file checks for several people, call this tool once per person. filed_at is the date on the REPORT, not today. It is required: do not default it to today, because a wrong completion date silently misdates a compliance record and is indistinguishable afterwards from one the caller chose. If the user did not say, ask. personnel accepts an id, email, or name -- resolved the same way other personnel tools resolve people. If resolution needs a decision (ambiguous name, multiple workspaces), the response carries error/action_required with candidates -- ask the user, do not guess or retry with a different value. NOT idempotent -- do not blindly retry. Repeating a call that already succeeded creates a SECOND Background Check record for the same person and date. If a call times out or its result has outcome_unknown, check whether the check already exists before calling again. Returns the created Background Check record, or an error dict on failure.

Drata_createControl

ChatGPT
Create a new compliance control in a workspace. WRITE operation that creates new tenant data. Requires the API key to have the 'Controls: Create Control' (controls-post) permission -- different from controls-put used by Drata_updateControl. Required: name, description, code (must be unique within the workspace; the API returns 400 with message DUPLICATE_CONTROL_CODE on duplicate). 🚫 NEVER auto-generate `name`, `description`, or `code` -- if the user has not explicitly provided any of these, you MUST ask for them before calling this tool. Optional metadata: question, activity, custom_fields. Optional relations -- prefer augmented inputs over raw IDs: owners (each entry auto-resolved: numeric ID → email → first/last name → partial), policy_names (auto-resolved by name -- preferred over policy_ids; resolves via GET /policies?name=; aborts if any name is ambiguous or not found), requirement_ids (use Drata_listRequirements first to look up IDs by code/name). Numeric ID escapes: policy_ids. Other: report_ids, test_ids. Evidence linking always happens AFTER control creation: to link an existing evidence library item, call this tool first to get the new control's code, then call Drata_updateEvidence with that code in control_codes. To upload a new evidence file, use Drata_createEvidence after creation. This tool does NOT attach evidence directly. Returns the created control object (with server-assigned id) on 201, or an error dict on failure (403 = missing permission, 400 = validation OR duplicate code (detail.message will be DUPLICATE_CONTROL_CODE), 404 = workspace not found).

Drata_createEvidence

ChatGPT
Register a new evidence bucket in Drata's reusable Evidence Library. WRITE operation. Requires the API key to have the 'Evidence: Create' (evidence-post) permission. Evidence items are reusable across controls -- register once, link to many. Vocabulary: an evidence bucket is a container -- users call it a folder or a bucket. It holds one or more files (also documents, attachments, artifacts). This tool creates the container; to change what's inside one that already exists, use Drata_updateEvidenceArtifacts. Choosing the artifact source: 1. The user wants to put FILES in it: set upload_files=True on THIS call -- one link, and the bucket and its files are created together when they Confirm. Do NOT create the bucket empty and then add files as a second step: that leaves an empty bucket behind if they never finish. 'Create an evidence bucket and upload two screenshots' is ONE call. 2. The evidence is a link, not a file (a URL or a Jira ticket was named): pass url or ticket_url. No upload of any kind. 3. Nothing to put in it yet: pass none of the above and you get an empty bucket, created immediately. Supported content sources -- provide AT MOST ONE of: upload_files=True (browser link the user drags files onto), url (link to evidence hosted elsewhere), or ticket_url (Jira ticket URL; API regex-validates format). On the upload_files path the tool returns a browser upload link and the user drags their files in. The user can add as many files as they want on that page -- there is no need to call this tool once per file, and no way to attach a second file by calling it again (that would create a second evidence bucket). File bytes never pass through Claude or this tool, so uploading always needs the user. File uploads create nothing until the user confirms: the evidence bucket and its files are written together when the user presses Confirm on the upload page. If they abandon it, no evidence bucket is created -- so do not tell the user the item exists until they say the upload finished, and do not immediately call Drata_listEvidence expecting to find it. File-upload flow: when the response contains a top-level user_message field, you MUST relay it VERBATIM to the user as plain text (NOT inside a code block, NOT in backticks). Do NOT add disclaimers, do NOT suggest alternative paths (no 'or upload in the Drata UI', no curl commands). Above the message, briefly say what will happen (e.g., 'Once you upload, this will create the "Pen Test Report" evidence bucket'). Below the message, do not add anything. Required fields depend on whether the bucket starts with an artifact: - Empty bucket (no upload_files, no url, no ticket_url): only owner is required. Do NOT ask for or send a renewal schedule -- renewal describes an artifact, and the API REJECTS renewal fields on an empty bucket. Just create it. - With an artifact: owner AND renewal_schedule_type are required, plus renewal_date when that is CUSTOM. Ask for the missing ones in ONE message before calling. filed_at auto-defaults to today and does not apply to file uploads -- each file is filed under the date it is uploaded. Returns the created evidence bucket with its server-assigned id for url and ticket_url sources; for file uploads, returns the upload link instead and the item is created on confirm.

Drata_createRisk

ChatGPT
Create a new custom risk in a risk register. Risks are potential threats to the organization's security, reputation, or financial health. title and description are required and must come from the user -- never invent them. treatment_details, anticipated_completion_date, completion_date, residual_impact, and residual_likelihood are only stored when treatment_plan is set to something other than UNTREATED -- the API silently drops them otherwise rather than rejecting the call, so omit them yourself when leaving treatment_plan UNTREATED or unset. owners and reviewers accept names, emails, or numeric IDs and are resolved server-side; unresolved or ambiguous entries abort the call rather than guessing. Only users with an admin/risk-management role can be assigned -- others are silently dropped (not rejected), so always check the returned owners/reviewers against what you requested.

Drata_createVendor

ChatGPT
Create a new third-party vendor for the account (TPRM). Vendors are account-scoped (no workspace ID). Only name is required; all other attributes are optional. Set owner (user id, email, or name) for the person responsible for vendor compliance. Returns the created vendor including its id.

Drata_deleteEvidence

ChatGPT
Permanently delete an ENTIRE evidence bucket -- the folder AND every file in it, plus all version history and control linkages. ⚠️ Destructive -- this cannot be undone. Deleting ONE file instead? Use Drata_deleteEvidenceArtifact. Users say 'folder' or 'bucket' for the evidence bucket and 'file', 'document' or 'attachment' for what's inside it. 'Delete the pen test folder' means this tool; 'delete the scan from the pen test folder' means Drata_deleteEvidenceArtifact. When it is ambiguous which one they mean, ask -- the difference is losing one file versus losing everything. 🔒 Two-step confirmation required -- never skip this: 1. Call with confirm=False (default) first -- returns a preview without touching data. 2. Show the user the preview and ask explicitly for approval. 3. Only call again with confirm=True after receiving explicit user approval. Use Drata_listEvidence to look up an item's numeric ID before deleting. Requires the delete:evidence OAuth scope.

Drata_deleteEvidenceArtifact

ChatGPT
Permanently delete ONE file from an evidence bucket, including that file's entire version history. The evidence bucket itself and its other files are kept. ⚠️ Destructive -- this cannot be undone. Deleting the WHOLE evidence bucket? Use Drata_deleteEvidence -- that removes the folder and every file in it. Users say 'folder' or 'bucket' for the evidence item and 'file', 'document' or 'attachment' for what's inside. 'Delete the scan from the pen test folder' means this tool; 'delete the pen test folder' means Drata_deleteEvidence. When it is ambiguous, ask. Prefer archiving: if the user wants the file out of the way but recoverable, Drata_updateEvidenceArtifacts with archive=['<file>'] is reversible; this is not. 🔒 Two-step confirmation required -- never skip this: 1. Call with confirm=False (default) first -- returns a preview without touching data. 2. Show the user the preview and ask explicitly for approval. 3. Only call again with confirm=True after receiving explicit user approval. Requires the delete:evidence OAuth scope.

Drata_deleteRisk

ChatGPT
Permanently delete a single risk from a risk register. ⚠️ Destructive -- this cannot be undone. 🔒 Two-step confirmation required -- never skip this: 1. Call with confirm=False (default) first -- returns a preview without touching data. 2. Show the user the preview and ask explicitly for approval. 3. Only call again with confirm=True after receiving explicit user approval. risk_id accepts either the numeric ID or the risk's display code (e.g. 'CR-08'). You do NOT need to resolve risk_register_id first on a multi-register account -- if omitted, the tool locates which register actually contains this risk and uses it silently; it only asks you to pick when the risk id genuinely exists in more than one register, or returns a clear error if it exists in none. Requires the delete:risk OAuth scope.

Drata_deleteVendor

ChatGPT
Permanently delete a vendor from the account (account-scoped; no workspace ID). Provide the vendor_id (from Drata_listVendors). This removes the vendor and its associated data and cannot be undone. 🔒 Two-step confirmation required -- never skip this: 1. Call with confirm=False (default) first -- returns a preview and touches no data. 2. Show the user what will be deleted and get explicit approval. 3. Only call again with confirm=True after the user approves.

Drata_excludeMonitoringTestResources

ChatGPT
Exclude specific failing items from one monitoring test, with a reason, so they stop counting against it. Works for cloud resources and for codebase Findings alike; the tool reads the test and sends each item to the right place. Target the items, not the test. Name the items by their id, resourceName or findingId from Drata_getMonitoringTestFindings, or pass connection_id to preview every current failure from one connection. With neither, nothing is excluded and the failures are returned as candidates to confirm with the user. Only current failures are excluded: an item that starts failing on a later run is not. Two-step. The first call returns a preview listing exactly which items would be excluded, and changes nothing. Obtain explicit user approval, then call again with confirm=True, passing the preview's targetIds as resources when the preview was by connection. A reason is required and is shown to auditors: record the user's own words, without extending or rewording them, and ask for a reason rather than reusing one given for other items. These are resource exclusions on one test; they are unrelated to the personnel exclusions in Drata_excludePersonnelFromComplianceChecks. Read the per-item outcome in the response. data[] has one entry per item, in request order, each with its own statusCode and outcome, and summary counts them. A SKIPPED_ outcome means nothing needed to change, while a FAILED_ outcome means that item needs another look. Report what actually changed, and report failures as failures. Use Drata_reincludeMonitoringTestResources to undo this. Use testId from Drata_searchMonitoringTests, not id.

Drata_excludePersonnelFromComplianceChecks

ChatGPT
Record that one or more named compliance checks do not apply to specific people, with a reason and an optional date range. Handles up to 500 people in a single call. This is the correct tool for 'this check does not apply to these people'. It is scoped to the checks you name and changes nothing else: every other compliance check still applies to them, their personnel status is unchanged, and they stay in the audit population. Do not use Drata_updatePersonnel to set someone OUT_OF_SCOPE for this purpose. That removes the person from every test and from the audit population, which is a much broader change than excluding them from one check and is almost never what is being asked for. Two-step. The first call returns a preview naming who, which checks, and the exact start and end instants, and changes nothing. Obtain explicit user approval, then call again with confirm=True. People may be given as numeric IDs, email addresses or full names. A reason is required and is shown to auditors, so pass the user's actual justification rather than a placeholder. Omit end_date for an indefinite exclusion. Read the per-person outcome in the response. successCount counts only people actually changed; skippedCount counts everyone else, whether skipped or failed. Read results[].outcome to tell them apart: a SKIPPED_ outcome means nothing needed to change (including a person outside this workspace's scope), while a FAILED_ outcome means that person needs another look. Report what actually changed, and report failures as failures. Use Drata_reincludePersonnelInComplianceChecks to undo this.

Drata_getCompany

ChatGPT
Get the account's own company information -- the organization's profile in Drata. This is the customer's OWN company, NOT a third-party vendor; for vendors use Drata_listVendors / Drata_getVendor instead. Account-scoped: takes no arguments and no workspace ID -- the company is resolved from the authenticated account. READ. Use this to answer questions about the organization itself: its name, legalName, description, logoUrl (the company logo / image URL), primary domain, and language. It also reports account configuration and compliance settings -- securityTraining, hipaaTraining, backgroundCheck, agentEnabled, manualUploadEnabled -- and the account entitlements (the plan and features enabled). Returns a single company object (also including accountId, createdAt, updatedAt). This tool is read-only -- it cannot change the company name, logo, description, or any setting.

Drata_getEvidence

ChatGPT
Get ONE evidence bucket by name or id. READ-only. Use this to answer questions the listing can't: when a bucket renews (renewalDate, renewalScheduleType) and what status it's in. Drata_listEvidence omits those fields. Renewal belongs to the bucket, not to individual files. If the user asks when a FILE expires, tell them files don't expire on their own -- the bucket carries the renewal date, and this tool reports it. For per-file dates (when a file was added, filed, or archived) use Drata_listEvidenceArtifacts. Requires the read:evidence OAuth scope.

Drata_getMonitoringTestFindings

ChatGPT
Read the individual items a monitoring test checked, for one test. A monitoring test reports a pass or fail for the whole test; this tool shows the individual cloud resources behind that result -- each with its provider, account, region and tags -- which is what remediation actually needs. Two buckets. result='failing' returns the items that did not satisfy the test. result='passing' returns the items that did. Some tests do not report passing items, so an empty passing page is not evidence that a fix failed -- check the failing bucket too. Excluded items are not failures. By default the failing bucket contains only genuine failures. Pass include_exclusions=True to also see items an administrator has deliberately excluded; each carries status EXCLUDED, without the reason. An excluded item is a decision already taken, not outstanding work -- never report one as a gap, and never propose remediating it without saying it is currently excluded. These are resource exclusions on the test, unrelated to the personnel exclusions in Drata_listComplianceCheckExclusions. Use testId from Drata_searchMonitoringTests, not id. Results are paginated; pagination.totalCount is returned on the first page.

Drata_getVendor

ChatGPT
Get a single vendor by ID, with optional expanded subcollections. Use the vendor id from Drata_listVendors.

Drata_listComplianceCheckExclusions

ChatGPT
List the compliance-check exclusions in force for a workspace. An exclusion records that one or more named compliance checks do not apply to one person, with a reason and an optional end date. Use this to see what has already been decided before proposing a new exclusion, and to explain why someone is not counted as failing a check. These are personnel exclusions; they are unrelated to the cloud-resource exclusions Drata_getMonitoringTestFindings reports as EXCLUDED. Results are cursor-paginated: pass pagination.cursor back as cursor for the next page. No total count is returned.

Drata_listDevices

ChatGPT
List a person's DEVICES and each device's per-monitor compliance checks -- the same view as the personnel 'Devices' tab in Drata. READ. Requires the 'Devices: Read' (devices-get) permission. Returns each device with a checks map of monitor -> status (PASS/FAIL/EXCLUDED) -- the same five requirements the device tab lists: PASSWORD_MANAGER, HDD_ENCRYPTION, ANTIVIRUS, AUTO_UPDATES, LOCK_SCREEN -- plus isDeviceCompliant. Use for 'show me <person>'s devices', 'is <person>'s laptop encrypted?', or 'what's failing on <person>'s machine?'. This reflects the RAW per-device monitor result (what the device reports). For the person's overall compliance POSTURE -- which also honors manually-uploaded evidence and can differ from the raw device result -- use Drata_searchPersonnelCompliance / Drata_searchDeviceCompliance. Identify the person by name (full/partial), email, or personnel id via person.

Drata_listEvidence

ChatGPT
List or search evidence buckets in Drata's Evidence Library. READ operation. Requires the 'Evidence: Read' (evidence-get) permission. Vocabulary: this lists the CONTAINERS -- users call them folders, buckets, or evidence buckets. It does not list what's inside them. 'Show me my pen test folders' is this tool; 'what's IN the pen test folder' is Drata_listEvidenceArtifacts, which lists that item's individual files (documents, attachments, artifacts). Native API filters (efficient, server-side): name (matches items whose name starts with this text), statuses (list of: NEEDS_ARTIFACT, ERROR, READY, EXPIRING_SOON, EXPIRED, NEEDS_SOURCE, NEEDS_ATTENTION), sort/sort_dir, cursor/size (1-500), expand, include_total_count. Augmented filters (client-side; auto-forces required expansions): control_code, control_id, owner (email / id / name). Multiple augmented filters are AND-combined. Tool paginates up to 10 pages. Returns paginated {data, pagination}. Augmented responses also include _augmentedFilter metadata.

Drata_listEvidenceArtifacts

ChatGPT
List the individual files inside ONE evidence bucket. READ-only. Vocabulary: an evidence bucket is the container -- users call it a folder, a bucket, or an evidence bucket. What's inside it are files (also called documents, attachments, or artifacts). This tool answers 'what's IN the pen test folder?'. To list the folders themselves, use Drata_listEvidence. Returns one entry per file, not one per version, so each id is stable across replacements and is the id the other file tools accept. Filter with name (prefix match), artifact_types, and is_archived (True = past files only, False = current only, omit for both). Counting: this endpoint returns no total count, so a 'how many files?' question needs the returned page (raise size if there may be more). Requires the read:evidence OAuth scope.

Drata_listPersonnel

ChatGPT
List or look up personnel -- employee/contractor compliance records, one per Drata user (distinct from Drata_listUsers, which returns auth identities). READ. Requires the 'Personnel: Read' (personnel-get) permission. Single-person lookup (exact, reflects very recent changes): pass personnel_id, email, or person (name/email, resolved automatically). Use for 'is John Doe compliant?' and to read back a record right after an update. WHY is someone non-compliant / their full check checklist: single-lookup with expand=['complianceChecks'] (or slim=True) returns every compliance check with its status -- both PERSON checks (accepted policies, BG check, MFA, training, offboarding) and DEVICE checks (antivirus, disk encryption, agent, etc.). slim's failingChecks lists exactly which checks failed, so you can say whether the failure was a device check or a personnel check like policy acceptance. Discovery / list (OpenSearch-backed): q (fuzzy name/email/job-title), employment_status, compliance_status (OVERALL PASS/FAIL/...), sync_status, group_ids, sort (createdAt/updatedAt/name/startDate), cursor/size, expand, facets, include_total_count. Eventually consistent -- a just-written record may lag a moment; read it back by email/id for certainty. Filter by a CUSTOM FIELD: accounts define their own personnel custom fields (e.g. 'Department', 'Manager', 'Category'). When the user names one, ALWAYS attempt the query -- do not assume the field doesn't exist or refuse; the name is resolved server-side and you get a clear error if it's genuinely unknown. Pass custom_field (the field NAME) plus custom_field_equals for an exact value, or custom_field_min/custom_field_max for an inclusive range (bounds are inclusive, so 'Tenure > 5' on an integer field is custom_field_min='6'). Answers 'who has <custom field> <comparison> <value>?'. For a SPECIFIC failing reason (e.g. who hasn't accepted policies, whose disk isn't encrypted) use Drata_searchPersonnelCompliance or Drata_searchDeviceCompliance. Use slim=True + size for large result sets; facets/include_total_count for 'how many'.

Drata_listPersonnelGroups

ChatGPT
List the personnel groups -- the teams / departments / identity-provider groups that personnel belong to (e.g. 'Engineering', 'Accounting'). READ. Requires the 'Personnel: Read' (personnel-get) permission. Use this to answer 'what groups/teams/departments do we have?', to show the user the available groups, or to disambiguate when a group name is unclear before filtering personnel by it. Filter by name with q (case-insensitive substring -- 'eng' matches 'Engineering'); omit q to list all. Returns paginated {data, pagination} with each group's id and name; paginate via cursor. Use include_total_count (with size=1) for 'how many groups?'. Once you have a group, filter people by it via the group parameter on Drata_listPersonnel, Drata_searchPersonnelCompliance, or Drata_searchDeviceCompliance (which accept a group NAME directly).

Drata_listPolicies

ChatGPT
List all account policies. Default choice for policy lookup. Works with both API key and OAuth. Supports optional name filtering (partial match). Use this to browse policies, count them, or resolve policy names to IDs for linking to controls. pagination.totalCount on the first page is the authoritative count -- read it directly instead of paginating to tally. Pass assigned_to_user (a user ID, email, or name) to list the policies ASSIGNED to that user instead of the full catalog. Use Drata_searchPolicies only for semantic/AI search over a specific user's assigned policies.

Drata_listRequirements

ChatGPT
List or search framework requirements in a workspace. READ operation. Use this to look up requirement IDs by code or name before calling Drata_createControl or Drata_updateControl with requirement_ids. Example: to find the ID for 'CCPA 1798.145', call with framework_tag=['CCPA'], code_or_name='1798.145'. Native API filters (server-side): framework_tag (single tag or list, e.g. 'SOC_2' or ['SOC_2', 'HIPAA']), framework_slug, is_in_scope, is_ready, expand (controls, customFields), include_total_count, cursor, size (1-500). Augmented filter (client-side): code_or_name -- paginates up to 10 pages and returns only requirements whose code or name contains the string (case-insensitive). Combine with framework_tag to narrow the search (strongly recommended -- without a tag filter this may scan thousands of requirements). Returns {data, pagination}. Each requirement includes id (numeric, for use in requirement_ids), code, name, and frameworkTag.

Drata_listRiskRegisters

ChatGPT
List all risk registers for the account. Risk registers are containers that organize and group related risks together. Most accounts have a default register (ID 1).

Drata_listTasks

ChatGPT
List Tasks in a workspace. Tasks are to-do items assigned to people to keep compliance work on schedule, such as renewing a policy, collecting evidence, or approving a control. Always include task IDs, titles, status, and due dates in responses. This returns only the Tasks people created: for Drata-assigned work (policy renewals, vendor reviews, and similar) or future occurrences of recurring Tasks, use Drata_listUpcomingTasks. Filter by status (PAST_DUE, INCOMPLETE, or COMPLETED; INCOMPLETE excludes PAST_DUE, so open tasks need INCOMPLETE then PAST_DUE), task_type (POLICY_RENEWALS, EVIDENCE, VENDOR, EXTERNAL_EVIDENCE, REPORT, GENERAL, CONTROL, RISK, CONTROL_APPROVALS, POLICY_APPROVALS), a due date range (due_date_from and due_date_to, YYYY-MM-DD, inclusive), or a title or description prefix. Only filter by assignee_id when the user asks about specific people, such as 'my tasks'. assignee_id and created_by_id take numeric User IDs: use Drata_lookupUserIdentity to turn a name or email into one. control_id takes a numeric Control ID from Drata_searchControls. Sort by createdAt, updatedAt, or dueDate. Expand assignee, createdBy, controls, risks, policies, or schedule to include those linked records. Returns paginated results (default 50, max 500). If pagination.cursor is set, the list is partial: say so (for example, 'the first 50'), and offer to fetch more instead of presenting it as complete. Pass include_total_count=True on the first page to report '50 of N' from pagination.totalCount. 🔒 Requires Tasks: List and Get Tasks permission.

Drata_listUpcomingTasks

ChatGPT
List Upcoming Tasks in a workspace: every to-do item with a due date, from every source. Unlike Drata_listTasks, which returns only the Tasks people created, this also includes Drata-assigned work from policy renewals, vendor reviews, external evidence due dates, library document renewals, and control approvals, plus future occurrences of recurring Tasks. Always include each item's name, task type, due date, and status in responses. Filter by status (PAST_DUE, INCOMPLETE, or COMPLETED; INCOMPLETE includes PAST_DUE), task_type (POLICY_RENEWALS, EVIDENCE, VENDOR, EXTERNAL_EVIDENCE, REPORT, GENERAL, CONTROL, RISK, CONTROL_APPROVALS, POLICY_APPROVALS), or a due date range (due_date_from and due_date_to, YYYY-MM-DD, inclusive). Without due_date_to, recurring occurrences are projected six months ahead. For work that is coming up, use status INCOMPLETE with due_date_from set to today and a due_date_to: when the user names no window, use about four months ahead. For open work including overdue, use INCOMPLETE with no due_date_from. Don't call PAST_DUE for questions about upcoming or next work: overdue Tasks aren't coming up. Call it only when the user asks about overdue or late work. Otherwise, offer to check overdue work in one line. Sort by dueDate. A recurring occurrence with no Task created for it yet has a null sourceId and a configurationId: to create it, pass that configuration_id and the row's dueDate as due_date to Drata_createTask. Vendor and policy renewal tasks are account-level, so they appear in every workspace. Returns paginated results (default 50, max 500). If pagination.cursor is set, the list is partial: say so (for example, 'the first 50'), and offer to fetch more instead of presenting it as complete. Pass include_total_count=True with a due_date_to to report '50 of N': without due_date_to, pagination.totalCount is left out. 🔒 Requires Tasks: List and Get Tasks permission.

Drata_listVendorDocuments

ChatGPT
List the documents attached to a vendor (e.g. SOC reports, compliance reports, bridge letters, questionnaire attachments). Requires the vendor id (from Drata_listVendors). Filter by document type. To obtain a downloadable link for a document, set expand=['downloadUrl'] -- this returns a presigned URL that expires ~60 seconds after the response, so use it immediately and do not cache or store it; only request it when a download is actually needed. Returns paginated results with pagination.totalCount.

Drata_listVendorSecurityReviews

ChatGPT
List the security reviews for a vendor. Security reviews capture the assessment of a vendor's security posture (SECURITY questionnaires, SOC_REPORT reviews, uploaded reports) and their decision (PENDING, APPROVED, APPROVED_WITH_CONDITIONS, REJECTED). Requires the vendor id (from Drata_listVendors). Filter by status, type, or decision. Returns paginated results with pagination.totalCount on the first page.

Drata_listVendors

ChatGPT
List third-party vendors for the account. Fallback for Drata_searchVendors: use this only when Drata_searchVendors is unavailable or returns an error, since search covers every filter here and more. Vendors are external providers tracked for third-party risk management (TPRM). Always include vendor IDs in responses. Vendors are account-scoped (no workspace ID needed). There is no natural-language search for vendors -- narrow results using the filters (category, risk, type, status, impactLevel, renewalDate, renewalScheduleType). Filter by owner (security owner -- user id, email, or name) to find the vendors a person is responsible for. Accepts one person OR several (pass a list): multiple owners are OR'd into a SINGLE call (the union of their vendors), matching the app's security-owner filter -- do NOT issue one call per person. pagination.totalCount reflects the owner-filtered count. Listing vendors: with no status, the rows include PROSPECTIVE and ARCHIVED vendors. When the user asks generally for their vendors, list the current book in ONE call by passing status = CURRENT. Include PROSPECTIVE when the user asks about vendors under evaluation, and label them as prospective. Returns paginated results with pagination.totalCount on the first page; paginate via the cursor. The total and the breakdown below come back unasked, so this tool takes no count flags. Counting vendors (match the Drata UI): when NO status is filtered, the response includes a breakdown object -- {current, total, prospective, archived} -- computed for you, where current = total - prospective - archived, all respecting the ACTIVE filters (owner/risk/category/...). ALWAYS report breakdown.current as the headline count -- that is what the app shows by default -- and include the total and prospective alongside it (e.g. "412 current; 483 total, 51 prospective"). Do NOT report the raw total as the headline, and do NOT treat "current" as status=ACTIVE (ACTIVE is only one of several current statuses). The breakdown reflects the current filters, so with owner=X it is that person's current/total/prospective (the owner filter returns ONLY their vendors -- it does NOT mean they own every vendor), and with risk=HIGH it is current high-risk vs total high-risk. When the user NAMES a status, filter by it directly (no breakdown is added); "total/all" = pagination.totalCount. Use expand to include subcollections such as latestSecurityReviews or documents. To fetch a vendor's full security reviews or documents, take the vendor id from this tool and call Drata_listVendorSecurityReviews or Drata_listVendorDocuments.

Drata_listWorkspaces

ChatGPT
List all workspaces accessible to the authenticated user. Good for testing API connectivity.

Drata_lookupUserIdentity

ChatGPT
Look up a Drata account (login identity) and its roles / SSO identities / background-check documents. READ. Requires the 'Users: List Users' (users-get) permission. Always use this tool for ANY 'who is X' / 'what is X's role' question about a person in THIS Drata account -- including vague, abbreviated, or possibly-misspelled names ('Jhn Doe', 'John D'). This account's data is the source of truth: do NOT answer from prior knowledge or a web search, and do NOT fabricate an email/ID. If the name matches several people, the tool returns candidates -- present them and let the user pick. Use this tool for: 'what is X's role?' (expand=['roles']), SSO identities (expand=['identities']), account existence, or looking up someone who is NOT personnel -- an admin, service account, or external/guest user. Users are a SUPERSET of personnel, so this is the right tool when a person isn't found in personnel. Do NOT use this for employee compliance status, failing checks, or employment records -- use Drata_listPersonnel / Drata_searchPersonnelCompliance / Drata_searchDeviceCompliance for those. Identify by: user_id (exact numeric) · email (exact, fast) · name (full or partial, and FUZZY -- 'John Doe', 'John D', or a minor misspelling all resolve via personnel search; returns a candidate list when several match) · first_name/last_name (partial). When the user gives a name, pass it verbatim as `name` -- do NOT guess or fabricate an email/ID; the tool looks it up (fuzzy personnel search first, then the /users directory for non-personnel) and returns candidates if more than one matches. Returns paginated {data, pagination}; a user_id/email hit returns the single matching account.

Drata_reincludeMonitoringTestResources

ChatGPT
Remove exclusions from one monitoring test, so those items count against it again. This is the inverse of Drata_excludeMonitoringTestResources, for cloud resources and codebase Findings alike. Name the excluded items by resource id, name or findingId, or pass exclusion_ids when they are known. With neither, nothing is removed and the test's exclusions are returned as candidates to confirm with the user. Those items may start failing the test again from its next run, which can change the workspace's compliance posture, so confirm the user wants that before proceeding. Two-step. The first call returns a preview listing exactly which exclusions would be removed, and changes nothing. Obtain explicit user approval, then call again with confirm=True. Read the per-item outcome in the response. data[] has one entry per item, in request order, each with its own statusCode and outcome, and summary counts them. A SKIPPED_ outcome means nothing needed to change, while a FAILED_ outcome means that item needs another look. Report what actually changed, and report failures as failures. Use testId from Drata_searchMonitoringTests, not id.

Drata_reincludePersonnelInComplianceChecks

ChatGPT
Remove compliance-check exclusions, bringing people back into scope for the checks you name. This is the inverse of Drata_excludePersonnelFromComplianceChecks. Use it when an exclusion was created in error, or when the circumstance behind it has ended. An exclusion that also covers other checks is narrowed to those other checks rather than removed. Those people may start failing these checks again, which can change the workspace's compliance posture, so confirm the user wants that before proceeding. Two-step. The first call returns a preview naming who and which checks, and changes nothing. Obtain explicit user approval, then call again with confirm=True. Read the per-person outcome in the response. successCount counts only people actually changed; skippedCount counts everyone else, whether skipped or failed. Read results[].outcome to tell them apart: a SKIPPED_ outcome means nothing needed to change (including a person outside this workspace's scope), while a FAILED_ outcome means that person needs another look. Report what actually changed, and report failures as failures. Use Drata_listComplianceCheckExclusions first if you are unsure which exclusions exist.

Drata_retestMonitoringTest

ChatGPT
Re-run one monitoring test immediately instead of waiting for its next scheduled run. Use this to prove a remediation worked, after the underlying problem has actually been fixed. Two-step. The first call returns a preview and changes nothing. Obtain explicit user approval, then call again with confirm=True. A re-run has real cost. It performs the same cloud and connector work as a scheduled run. Do not call it in a loop, and do not call it to poll for a result. Accepted, not finished. A successful response means the run started and returns checkStatus: TESTING. To see the outcome later, find the test again with Drata_searchMonitoringTests (search by its name) or read its failing items with Drata_getMonitoringTestFindings once it has finished. If a run is already in flight the call is refused rather than queued, and if a run completed moments ago the test is briefly in a cooldown window and the call is refused too. In both cases wait and try later rather than retrying immediately. Use testId from Drata_searchMonitoringTests, not id.

Drata_searchControls

ChatGPT
Search or list compliance controls for a workspace. Controls are security and compliance measures that organizations must meet. Always include control codes in responses. Two types of operation: 1. List (no query): Returns paginated controls with filtering options. Use this for counting, reporting, or any question about how many controls match a condition (e.g. 'how many controls are not ready?', 'how many controls have no evidence?'). Use is_ready=False -- never use a search query -- when the user asks about not-ready OR 'failing' controls (a 'failing' control means it is NOT READY). Use is_ready=True for ready controls. Use has_passing_test=False ONLY when the user explicitly asks about failing/failed TESTS -- not for a generic 'failing control'. pagination.totalCount on the first page is the authoritative count. 2. Search (with query): Performs AI-powered semantic or keyword search. Use this only to find specific controls by name or topic (e.g. 'find controls about encryption'). Never use search mode for counting or status-based filtering -- it does not respect is_ready and returns approximate AI-ranked results, not an exact filtered set.

Drata_searchDeviceCompliance

ChatGPT
Find personnel by a SPECIFIC device-level compliance reason -- agent installed, password manager, disk encryption, antivirus, auto-updates, or lock screen. READ. Requires the 'Personnel: Read' (personnel-get) permission. Each filter takes statuses (usually ['FAIL']). Example: disk_encryption=['FAIL'] answers 'whose laptop isn't encrypted?'. Multiple filters AND together. Use facets for grouped counts without fetching rows, and employment_status to scope to current staff. For person checks (policies, training, BG check) use Drata_searchPersonnelCompliance; for overall status or plain lookup use Drata_listPersonnel.

Drata_searchMonitoringTests

ChatGPT
Search or list monitoring tests for a workspace. Monitoring tests are automated checks that run to verify control compliance. IMPORTANT - Two different IDs in response: testId is the test template ID (identifies the test type). id is the monitoring test instance ID (unique per workspace). Always use testId when referencing tests in responses. Two types of operation: 1. List (no query): Returns paginated tests with filtering options. Use this for comprehensive reporting, analysis, or when you need broad coverage of tests. Supports pagination, sorting, and multiple filters (result status, system status, type, source). For thorough analysis, paginate through results when needed. Returns pagination.totalCount on the first page for planning pagination. 2. Search (with query): Performs AI-powered semantic or keyword search. Use this to find specific tests matching a natural language query. Pagination and filter parameters are ignored in this mode. Analysis tips: Expand controls to inform prioritization of tests. Tests mapped to more controls or critical controls should be prioritized for remediation.

Drata_searchPersonnelCompliance

ChatGPT
Find personnel by a SPECIFIC person-level compliance reason -- accepted policies, background check, identity MFA, security/HIPAA/NIST-AI training, or offboarding. READ. Requires the 'Personnel: Read' (personnel-get) permission. Each filter takes statuses (usually ['FAIL']). Example: accepted_policies=['FAIL'] answers 'who hasn't accepted their policies?'. Multiple filters AND together. Use facets for grouped counts (e.g. how many FAIL vs PASS) without fetching rows, and employment_status to scope to current staff. For DEVICE checks (antivirus, encryption, etc.) use Drata_searchDeviceCompliance; for overall status or plain lookup use Drata_listPersonnel.

Drata_searchPolicies

ChatGPT
AI search over the policies ASSIGNED TO YOU (the authenticated user), using semantic or keyword search -- requires a search query. Scoped to your assigned policies, not the full account catalog. Policies are formal documents that define organizational rules and requirements for compliance. Always include policy names in responses. Returns policyId, name, description, and matching text chunks for each result. Requires OAuth auth -- not available with API key. Use Drata_listPolicies to browse/filter the full account catalog, or Drata_listPolicies with assigned_to_user to list (not search) a user's assigned policies.

Drata_searchRisks

ChatGPT
Search or list risks from the risk register. Risks are potential threats to the organization's security, reputation, or financial health. Always include risk IDs. Two types of operation, selected by whether `query` is set: 1. Structured list/filter (no query, OpenSearch-backed): rich filtering (status, treatment_plan, type, inherent/residual score ranges, anticipated-completion/completed date ranges, categories, controls, owners, reviewers, vendor_names, one custom field), 9 sort keys (residualScore default, highest risk first), facets (grouped counts for status/treatmentPlan/type/categories/controls/ownerEmails/reviewerEmails/vendorNames without fetching every row), and q for a fuzzy keyword search across title/description/riskId/vendorName that combines with every filter. Use this for comprehensive reporting, analysis, or broad coverage of risks. Returns pagination.totalCount on the first page for planning pagination. 2. Natural-language search (query set): AI-powered semantic or keyword search for a specific risk matching a free-form question. Only search_mode, size, expand, status, treatment_plan, min_score, max_score apply in this mode -- every other filter/sort/facet parameter is ignored. Structured mode searches across every risk register the caller can access by default -- on a multi-register account, omitting risk_register_id no longer asks which register to use; it searches all of them and each result is labeled with its register name (riskRegisterName). Pass risk_register_id (or workspace_id) to narrow to one or a subset instead. Natural-language `query` mode is still scoped to ONE risk_register_id -- on a multi-register account with none named/known, it asks which register to use; it does not search across registers. A place name is NOT a search term, and NOT a name-matching exercise. If the user asks about risks 'in <place>' and <place> could be a workspace (e.g. 'risks in Cloud Solutions'), do NOT pass it as q/query -- a keyword/semantic search only looks at risk title/description text, never at which workspace a risk's register belongs to, so it will come back empty and look like 'no risks exist' even when there are plenty. Also do NOT guess a risk_register_id because its NAME happens to share a word with <place> (e.g. picking a register called 'Cloud Finance' for a workspace called 'Cloud Solutions' just because both say 'Cloud') -- register names and workspace names are chosen independently and a shared word is a coincidence, not a verified mapping; that guess can silently point at the wrong register. The only reliable path: call Drata_listWorkspaces first, and if <place> matches a real workspace, resolve it and pass workspace_id (see its description) so the right register is used from the real register-to-workspace mapping, not from a name guess. Analysis tips: To prioritize risks, filter by 'ACTIVE' status and expand with 'controls'. A risk 'Needs attention' if any mapped control is not ready. 'Needs attention' is separate from treatment plan--it reflects control readiness, not risk strategy.

Drata_searchVendors

ChatGPT
Search third-party vendors, backed by OpenSearch. Vendors are external providers tracked for third-party risk management (TPRM). Always include vendor IDs in responses. Vendors are account-scoped (no workspace ID needed). Prefer this over Drata_listVendors for any vendor question. It is the only vendor tool with free-text search (q), custom-field filtering, renewal DATE RANGES, security-review and questionnaire filters, and facets. q is a fuzzy, typo-tolerant keyword search over the vendor name and servicesProvided that combines with every filter. Use it for 'vendors doing X'. It does NOT search notes or documents. Filter by owner (security owner: user id, email, or name) to find the vendors a person is responsible for; multiple owners are OR'd into a SINGLE call, so do NOT issue one call per person. Filter by ONE custom field via custom_field (its NAME) plus one of custom_field_equals / custom_field_min / custom_field_max. The name is resolved server-side, case-insensitively, so pass what the user said verbatim; an unknown or ambiguous name returns a clear error rather than failing silently. Ranges are INCLUSIVE, so express 'more than 5' as custom_field_min=6. Pass expand[]=customFields to see each matched vendor's values. Use facets to get grouped counts (status, category, vendorTypeId, risk, impactLevel, isSubProcessor, securityReviewStatus, securityReviewDecision, scheduledQuestionnaireStatus) without fetching every row. Facets and pagination.totalCount are first-page only; paginate via cursor. When you only need COUNTS, pass size=1 together with facets: every row in data is a full vendor record, so a large size wastes most of the response on rows you will not read. "Which vendors need attention?" maps to action_required=true. That is Drata's own definition (a current vendor, with a directory entry and an owner, whose review is due soon or overdue) and it is the authoritative answer to that question. If it returns zero, the answer IS "nothing currently needs attention": report that and STOP. Do NOT treat an empty result as a failed query and do NOT substitute a hand-rolled definition built from status, risk, or review coverage unless the user explicitly asks for one. If the response contains a breakdown object, report breakdown.current as the headline count and include total and prospective alongside it (e.g. "412 current; 483 total, 51 prospective"). Do NOT headline the raw total, and do NOT treat "current" as status=ACTIVE; ACTIVE is only one of several current statuses. Status defaults: omitting status returns every status EXCEPT PROSPECTIVE. ARCHIVED, FLAGGED, OFFBOARDED and all others are already included, so do NOT add a status filter hoping to widen the search; status only ever narrows. If a search returns too few results, the fix is a different q or fewer filters, not a status. To include prospective vendors pass status = CURRENT and PROSPECTIVE, because PROSPECTIVE alone returns only prospective vendors. Listing the vendor book: when the user asks generally for their vendors (no q, no status named), the default results include ARCHIVED vendors. List the current book instead by passing status = CURRENT, and do not present archived vendors as active. Include PROSPECTIVE when the user asks about vendors under evaluation, and label them as prospective. A q lookup for a specific vendor keeps the default, so an archived match is still found. Vendor type: vendor_type_ids filters the tenant-defined vendor-type taxonomy only. The legacy VENDOR/SUPPLIER/CONTRACTOR/PARTNER/OTHER enum is NOT filterable here; it is still returned on each vendor as type. If the user asks to filter by one of those legacy values, say it is not supported rather than substituting vendor_type_ids or filtering the page yourself. To fetch a vendor's full security reviews or documents, take the vendor id from this tool and call Drata_listVendorSecurityReviews or Drata_listVendorDocuments.

Drata_updateControl

ChatGPT
Update an existing compliance control in a workspace. WRITE operation. Requires the 'Controls: Update Control' (controls-put) permission. All fields are optional -- only send what you want to change. Identify the control via control_code (preferred, e.g. 'AC-1') or control_id (numeric). ⚠ Relation fields (owners, policy_names/policy_ids, report_ids, requirement_ids, test_ids) REPLACE the entire list -- they do NOT append. To ADD to a relation while keeping existing entries, first read the control's current values (e.g. via Drata_searchControls) and include them alongside the new ones; sending only the new value removes all others. Sending an empty list clears the relation; omitting a field leaves it unchanged. Prefer augmented inputs over raw IDs: owners (each entry resolved by ID / email / full / partial name -- most flexible), policy_names (resolved by name, combined with policy_ids). Returns the updated control on success, or an error dict on failure.

Drata_updateEvidence

ChatGPT
Update an evidence BUCKET's own details -- name, description, implementation guidance, owner, control mapping, renewal. WRITE operation. Requires the 'Evidence: Update' (evidence-put) permission. Identify the bucket via evidence_id (numeric) OR evidence_name (resolved internally). If evidence_name matches several, returns ambiguous_name with candidates. Vocabulary: the evidence bucket is the container -- users call it a folder or a bucket. This tool edits the CONTAINER only. ⚠ This tool does NOT touch files. To upload files, attach a link or ticket, replace, archive, or restore anything inside the bucket, use Drata_updateEvidenceArtifacts. It has no url, ticket_url, or file parameters -- sending one is an error, not a fallback. PUT semantics: only the fields you send are updated. Send renewal_schedule_type only when the user actually asks to change the renewal cadence -- adding files or links does not require one. ⚠ `controlIds` REPLACE quirk: providing control_codes or control_ids REPLACES the entire control mapping. To ADD a control, first call Drata_listEvidence with expand=[controls] to read current codes, then send the union. Returns the updated evidence bucket, or an error dict on failure.

Drata_updateEvidenceArtifacts

ChatGPT
Change the FILES inside an evidence bucket -- upload, attach links, replace, archive, restore. WRITE operation. Vocabulary: the evidence bucket is the container (folder/bucket); this tool acts on the files inside it (also called documents, attachments, artifacts). To rename the bucket or change its owner, renewal, or controls, use Drata_updateEvidence. What to set: - upload_files=True -- user wants to upload FILES. Returns a browser link; they can drop as many files as they want on that one page, so call this ONCE, not once per file. Nothing attaches until they press Confirm. - url / ticket_url -- attach ONE link or Jira ticket, applied immediately. The upload page is for files only, so never combine these with upload_files; make two calls if the user wants both. Attaching does NOT need a renewal schedule -- the bucket already carries one, so just attach it; don't stop to ask. - add_artifacts -- attach MANY links/tickets at once (e.g. every ticket in a list the user pasted). - replace_file -- swap one existing file's contents, keeping its id and history. Pair with upload_files=True or with url/ticket_url. - archive / restore -- move files between current and past, by name or id. Both are REVERSIBLE and keep version history. Adding a new file does NOT auto-archive the existing ones; archive them explicitly if that's the intent. Not this tool: to remove a file permanently use Drata_deleteEvidenceArtifact; prefer archive when the user may want it back. There is no metadata-only edit for a file -- renaming one or correcting its filing date means supplying its contents again via replace_file. Requires the update:evidence OAuth scope.

Drata_updatePersonnel

ChatGPT
Update a person's employment record in Drata. WRITE operation. Requires the 'Personnel: Update' (personnel-put) permission. Identify the person via `personnel_id` (numeric), `email`, or `person` (name/email, resolved automatically). Precedence: personnel_id > email > person. Only the fields you send are updated; omitted fields are left unchanged. Updatable fields: • started_at -- ISO 8601 start date. ⚠️ Manually setting this DISABLES automatic IDP/HRIS sync for that field on this person; tell the user when you do this. • separated_at -- ISO 8601 separation date; must be after started_at. • employment_status -- CURRENT_EMPLOYEE, FORMER_EMPLOYEE, CURRENT_CONTRACTOR, FORMER_CONTRACTOR, FUTURE_HIRE, UNKNOWN, OUT_OF_SCOPE, SERVICE_ACCOUNT, SPECIAL_FORMER_EMPLOYEE, SPECIAL_FORMER_CONTRACTOR. • not_human_reason -- free text. REQUIRED when employment_status is OUT_OF_SCOPE. Ask the user for it and send it in the SAME call. • custom_fields -- list of {id?|name?, value} objects (requires the Custom Fields & Formulas feature). Returns the updated personnel record, or an error dict on failure (404 = not found, 400 = validation, 403 = missing permission).

Drata_updateRisk

ChatGPT
Update an existing risk. risk_id accepts either the numeric ID or the risk's display code (e.g. 'CR-08', 'RISK-001'). You do NOT need to resolve risk_register_id first on a multi-register account -- if omitted, the tool locates which register actually contains this risk and uses it silently; it only asks you to pick when the risk id genuinely exists in more than one register, or returns a clear error if it exists in none. Omitted fields are left unchanged. owners, reviewers, category_ids, and control_ids REPLACE the entire relation when provided -- they do not append. To add to the existing set, read the risk's current values first (e.g. via Drata_searchRisks with expand) and include them alongside the new ones. Omit a relation to leave it unchanged; pass an empty list to clear it. treatment_details, anticipated_completion_date, completion_date, residual_impact, and residual_likelihood are only stored when treatment_plan is set to something other than UNTREATED -- the API silently drops them otherwise rather than rejecting the call. owners/reviewers accept names, emails, or numeric IDs and are resolved server-side; unresolved or ambiguous entries abort the call. Only users with an admin/risk-management role can be assigned -- others are silently dropped (not rejected), so always check the returned owners/reviewers against what you requested.

Drata_updateTask

ChatGPT
Update an existing Task in a workspace. Tasks are to-do items assigned to people to keep compliance work on schedule, such as renewing a policy, collecting evidence, or approving a control. Returns the updated Task. Always include the task ID, title, and due date in responses. task_id takes the numeric Task ID or the Task's title. Only the fields you pass change; omitted fields stay as they are. due_date is YYYY-MM-DD. assignee_id takes a numeric User ID: use Drata_lookupUserIdentity to turn a name or email into one. Set clear_assignee_id=True to unassign the Task, or clear_description=True to remove its description. control_ids, risk_ids, and policy_ids replace every existing link of that kind, and [] removes them all. To add or remove one link, first read the Task's current links with Drata_listTasks (expand controls, risks, or policies), then pass the full new list. Get the IDs from Drata_searchControls, Drata_searchRisks, and Drata_searchPolicies. This tool can't change a Task's status or mark it complete. 🔒 Requires Tasks: Update Task permission.

Drata_updateVendor

ChatGPT
Update an existing vendor (account-scoped; no workspace ID). Provide the vendor_id (from Drata_listVendors) plus only the fields you want to change -- this is a partial update, so omitted fields are left unchanged. Use status for lifecycle changes (e.g. ACTIVE, ARCHIVED). Set owner or relationship_contact by user id, email, or name (resolved automatically).

Drata_uploadPersonnelDocument

ChatGPT
File personnel compliance documents for one person -- security or HIPAA training evidence, MFA evidence, or offboarding evidence. WRITE operation. This is for documents about a PERSON's own compliance; it is not the tool for evidence attached to a control or an evidence library (use the evidence tools), nor for background checks, which are a separate record type. Requires the 'Users: Post Document' (users-post-document) permission / create:user-document OAuth scope. It also requires an Evidence scope (create or update): POST /upload-urls is gated on an Evidence ability regardless of purpose, so an app holding only create:user-document is refused before its own scope is checked. `OFFBOARDING_EVIDENCE` needs one more scope than the other three: create-restricted:personnel, because it is filed ABOUT a departing person rather than by them. Without it that type alone is refused while training and MFA evidence still work -- relay that as needing the restricted-personnel grant, not as a general permission failure. ONE document type per call. document_type is a single value, so filing a training certificate AND offboarding evidence for the same person is two calls. Do not try to encode two types in one call. Upload flow -- ONE call. You get back a user_message containing a link. Relay that message VERBATIM as plain text, not in a code block, so the link stays clickable. The link opens a Drata-hosted page where the user drags in as many files as they need and presses Confirm; every file becomes its own document for this person and this type. Do NOT ask the user for filenames or file paths, and do not call this tool a second time to finish the upload -- there is no second call. File bytes never pass through Claude or this tool. Nothing is created until the user presses Confirm, and a document earns compliance credit only after antivirus clears the file. An infected or oversized file leaves that document QUARANTINED; the others are unaffected. If the user never opens the link, or never confirms, no document is created at all. personnel accepts an id, email, or name, resolved the same way as other personnel tools. Prefer an email or a numeric USER id -- the id other tools return, NOT a personnel id: a numeric value is resolved as a user, so a personnel id here files the document against a different person. A NAME is the only input that needs a workspace to resolve, and this tool has no workspace argument. An ambiguous name comes back as error: ambiguous_person with candidates -- ask the user which person they meant, then retry with that person's EMAIL. A name on a multi-workspace account comes back as action_required asking which workspace; there is no workspace argument to answer that with, so ask the user for the person's email instead and retry with it. Never guess a different name. This purpose is gated per tenant and off by default. An account without it gets a 400 whose message is exactly Unsupported upload purpose: personnel_document. ONLY for that message, relay 'not enabled for this account' and do not retry or try another purpose. Any OTHER 400 is a problem with the request itself -- relay its message as-is, because calling a malformed field an account-level outage sends the user to an admin for something they could have fixed themselves. Minting a second link does not duplicate anything -- an unused link simply expires. But do not mint one per file: a single link takes as many files as the user has. Returns upload.webUploadUrl plus the user_message to relay, or an error dict.

Drata_createControl

Claude

Drata_createEvidence

Claude

Drata_createRisk

Claude

Drata_createVendor

Claude

Drata_deleteEvidence

Claude

Drata_deleteEvidenceArtifact

Claude

Drata_deleteRisk

Claude

Drata_deleteVendor

Claude

Drata_getCompany

Claude

Drata_getEvidence

Claude

Drata_getVendor

Claude

Drata_listDevices

Claude

Drata_listEvidence

Claude

Drata_listEvidenceArtifacts

Claude

Drata_listPersonnel

Claude

Drata_listPersonnelGroups

Claude

Drata_listPolicies

Claude

Drata_listRequirements

Claude

Drata_listRiskRegisters

Claude

Drata_listVendorDocuments

Claude

Drata_listVendorSecurityReviews

Claude

Drata_listVendors

Claude

Drata_listWorkspaces

Claude

Drata_lookupUserIdentity

Claude

Drata_searchControls

Claude

Drata_searchDeviceCompliance

Claude

Drata_searchMonitoringTests

Claude

Drata_searchPersonnelCompliance

Claude

Drata_searchPolicies

Claude

Drata_searchRisks

Claude

Drata_updateControl

Claude

Drata_updateEvidence

Claude

Drata_updateEvidenceArtifacts

Claude

Drata_updatePersonnel

Claude

Drata_updateRisk

Claude

Drata_updateVendor

Claude

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/drata"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/drata)