MCP App Store
by KeenEthics
Wolfia icon

Wolfia MCP Server for ChatGPT

By Wolfia

Overview

Wolfia keeps a company's approved security, privacy and compliance knowledge in sync: policies, audit reports, trust center documents and past questionnaire answers. Connect it to ChatGPT and ChatGPT answers from that approved knowledge instead of guessing, cites the source of every answer, and acts with your own Wolfia role. With Wolfia in ChatGPT you can ask a security or compliance question and get a cited answer, find a policy, report or trust center document and get a download link, look up what you told a customer in past questionnaires, see which questionnaire answers still need attention, and edit, approve or regenerate them. Every action runs as the signed-in user, inside their organization, so ChatGPT only sees and changes what that user can see and change in Wolfia. Changes that are hard to undo wait for a Wolfia Admin to approve them in Wolfia. You need a Wolfia account in your organization. Sign in with your usual Wolfia login, SSO or email, and approve the consent screen that names ChatGPT as the destination.

How to connect

  1. ChatGPT: Open the Wolfia listing in the ChatGPT app directory and choose Connect. ChatGPT listing Wolfia exposes 111 tools here.

Tools

add_knowledge

ChatGPT
Submit a new fact to your organization's knowledge base so Wolfia uses it when answering questionnaires and questions. Provide a short title and the fact_text. Optionally add context, business-segment tag_ids, and an is_public_override shareability flag. A client_reference_id belongs to this one fact: a repeated call with it returns the existing fact, and fact_text in the response is the fact as stored. answer_triage_verdict_id is the verdict_id of one open knowledge update (a gap or card on the Knowledge updates page) that this fact fills; one call fills at most one update. is_approved true in the response means the update is closed. is_approved false means the fact needs approval: the update is listed with is_waiting_for_approval true and closes once the fact is approved, or, when is_waiting_for_approval stays false, this text was already rejected and the update stays open. Errors: 409 when the client_reference_id already holds a different fact (nothing is saved and the update stays open) or the update is already closed or waiting for approval. Requires an expert or admin role.

add_knowledge_bulk

ChatGPT
Submit up to 25 new facts to your organization's knowledge base in one call. Items carry no verdict id, so a fact added here fills no knowledge update and every knowledge update stays open. Each item needs a client_reference_id, a short title and the fact_text; context, business-segment tag_ids and is_public_override are optional. The client_reference_id must be the permanent id of the record the fact comes from in your own system (for example a Notion page id or a database row id): the same id every time that record is sent, and never shared by two different facts. Row numbers, list positions and counters such as 1, 2, 3 repeat across runs, and a repeated id is treated as a fact that was already added. Returns one result per item, in order: CREATED with the new source_id, DUPLICATE with the existing source_id when that client_reference_id was already added (nothing is changed, so a repeated call after a failure is safe, and that source_id is the id of the existing fact for changing its text), or FAILED with a failure_reason. One failed item does not stop the others. Requires an expert or admin role.

approve_pre_approved_answer

ChatGPT
Mark one pre-approved answer as approved. pre_approved_id is the id of a pre-approved answer (the pre-approved answer list with is_approved false lists unapproved answers); there is no body. Sets is_approved to true and records the caller as approver with approved_at in microseconds; an answer that is already approved is returned unchanged. When the organization requires approval before pre-approved answers are used, only approved and active answers are used in answering. Removing the approval undoes it. Returns the answer with the same fields as an item of the paginated pre-approved answer list. Errors: 404 when the id does not exist or the answer was deleted. Requires an expert or admin role.

approve_questionnaire_answers

ChatGPT
Approve or unapprove questions in one call. Takes question_ids and approve, true to approve and false to unapprove. Returns approved_count and message.

approve_trust_access_request

ChatGPT
Approves a Trust Center access request by its id (the access request id shown in the access request list) and emails the requester their access link. approval_scope is required: USER grants access to the requester only, DOMAIN grants access to everyone at the requester's email domain. When the portal requires a DocuSign or Ironclad NDA, the request moves to awaiting NDA signature and the link is sent after signing. Works on pending, rejected and NDA-declined requests and fails on any other status. The access token is never returned. Requires an expert or admin role.

create_internal_trust_access_request

ChatGPT
Creates a Trust Center access request for an external contact on behalf of an admin. The request is auto-approved and a tokenized invite link is emailed to the contact, matching the Grant access and send invite action in the admin UI. Accepts contact_name, email, reason, and company_name, which may be omitted only when linked_crm_provider and linked_crm_account_id name the contact's CRM account (the account is then named from that CRM account), plus optional document/category scopes (empty scopes = full portal access) and skip_nda. If the portal requires a DocuSign NDA and skip_nda is false, the invite is sent after the NDA is signed. Requires an expert or admin role.

create_pre_approved_answer

ChatGPT
Add a new pre-approved answer to the organization's library. Body: question (required, not empty); answer (optional, the company-wide default answer); category_ids (optional list of category ids, as they appear in category_ids on pre-approved answers); variants (optional list of answers for specific business segments, each with answer (required, not empty) and tags (required, one or more business segment ids; every ACTIVE segment id appears either in default_answer_tag_ids or in a variant's tags in the full detail of a pre-approved answer)). At least one of answer or variants is required. The new answer is active and not approved. Returns the created answer with the same fields as an item of the paginated pre-approved answer list, but its variants list is always empty in this response, because a variant whose business segments are not ACTIVE or overlap an earlier variant in the same request is skipped without an error; the saved variants appear in the full detail of the answer. A variant's internal_note is not saved here; it is set by editing the variant. Errors: 400 when neither answer nor variants is sent; 409 when an answer with the same question already exists, compared ignoring case and extra spaces; 422 when question is missing or a text field is empty. Requires an expert or admin role.

create_questionnaire_file_upload_url

ChatGPT
First step of uploading a questionnaire file (xlsx, xlsm, xls, docx, pdf, csv, png, jpg, or a zip of those files). Returns a presigned URL that accepts the raw file bytes with an HTTP PUT, plus a file_upload_id. Once the PUT succeeds, the file_upload_id is what questionnaire submission takes to create the questionnaire from that file. Unlike an inline base64 file_content argument, which can stall the tool call on large files, the file bytes never pass through a tool call.

create_user_correction

ChatGPT
Submit a correction that teaches Wolfia how a question should be answered, recording the corrected answer so future answers improve. Requires an expert or admin role.

create_variant

ChatGPT
Add an answer variant to one pre-approved answer: a different answer used instead of the company-wide default answer when Wolfia answers for one of the variant's business segments. pre_approved_id is the id of a pre-approved answer. Body: answer (required, not empty); tags (required, one or more business segment ids; each must be ACTIVE and not already in another variant of this answer; every ACTIVE segment id appears either in default_answer_tag_ids or in a variant's tags in the full detail of the answer, and the ones in default_answer_tag_ids are free); internal_note (optional note for reviewers that answering never uses). Returns the variant: id, pre_approved_answer_id, answer, internal_note, original_answer, tags, created_by, created_at and updated_at in microseconds, updated_by. Errors: 409 when the pre-approved answer does not exist or was deleted, a business segment is not ACTIVE, or a segment is already in another variant of this answer; 422 when answer or tags is missing or empty. Requires an expert or admin role.

deactivate_pre_approved_answer

ChatGPT
Stop Wolfia from using one pre-approved answer when it answers questionnaires and chat, without deleting it. pre_approved_id is the id of a pre-approved answer; there is no body. Sets is_active to false; an answer that is already inactive is returned unchanged. Reactivating the answer undoes it. Returns the answer with the same fields as an item of the paginated pre-approved answer list. Errors: 404 when the id does not exist or the answer was deleted. Requires the admin role.

delete_knowledge

ChatGPT
Delete a knowledge-base fact by its id so Wolfia stops using it when answering. Works for both a manual fact and an extracted fact (for example one extracted from a questionnaire answer, email, or uploaded document). Returns a 404 error if no fact has that id in this organization — it does NOT silently report success. Requires an expert or admin role.

delete_pre_approved_answer

ChatGPT
Delete one pre-approved answer by moving it to the trash. Wolfia stops using it in answering, it no longer appears in the pre-approved answer list or detail, and the list with deleted_only true lists it. The row and its variants are kept, so restoring it from the trash brings it back unchanged. pre_approved_id is the id of a pre-approved answer. A call without confirm_delete true is refused and deletes nothing; confirm_delete true states that the user confirmed deleting this answer (its question). Deactivating an answer, unlike deleting it, stops Wolfia using it while keeping it in the library. Returns success and message. Errors: 404 when the id does not exist or the answer is already deleted. Requires the admin role.

delete_variant

ChatGPT
Permanently delete one business segment variant of a pre-approved answer. After the delete, answering for those business segments no longer uses the variant's text and uses the company-wide default answer when there is one. This cannot be undone: restoring the pre-approved answer from the trash does not bring back a deleted variant, while editing a variant changes its text or segments and keeps it. pre_approved_id is the id of the pre-approved answer and variant_id is the id of an entry in its variants. A call without confirm_delete true is refused and deletes nothing; confirm_delete true states that the user confirmed deleting this variant (its answer and segments). Returns success and message. Errors: 404 when the variant does not exist, does not belong to pre_approved_id, its pre-approved answer was deleted, or it is the only variant of an answer that has no company-wide default answer (that answer needs a default answer before its last variant can be deleted). Requires the admin role.

deny_trust_access_request

ChatGPT
Denies a pending Trust Center access request by its id (the access request id shown in the access request list). The requester does not get access. Fails on any status other than pending. Requires an expert or admin role.

dismiss_knowledge_gap

ChatGPT
Dismiss one open knowledge update so it no longer appears on the Knowledge updates page. verdict_id is the verdict_id of an open gap or card on any tab of the Knowledge updates page (Missing information, Conflicts or What we learned). Dismissing fits a gap that needs no knowledge base change, for example the question is out of scope or the knowledge is already handled elsewhere. Dismissing changes no fact or answer; if the gap was waiting for approval of a fact saved from it, that approval request is rejected. Returns verdict_id and dismissed_at in microseconds, or a 404 error when the verdict_id does not exist. Requires permission to edit the knowledge base (an expert or admin role).

edit_questionnaire_answer

ChatGPT
Set the answers of one question's fields and optionally approve that question in the same call. Takes question_id; field_answers, a list of {response_id, answer}, where response_id is the id of one of the question's fields as listed with its answers; and approve, which is null to leave the approval state unchanged. A SINGLE_SELECT field's answer must be one of the options listed for that field. An edit that changes an answer clears the question's approval unless approve is true. Returns the question's is_approved, approval_cleared_by_edit, and the fields it saved.

export_questionnaire

ChatGPT
Start an asynchronous export of one questionnaire's answers and return its export_id. Takes questionnaire_id; mark_as_completed (required), whether this export also marks the questionnaire Completed in Wolfia, which the Wolfia web app does by default; output_format, one of docx, original, pdf, text, xlsx; and approval_filter, one of all, approved, awaiting_approval, where anything other than all is supported for the xlsx output format only. The export runs in the background; its status, and the file once it is complete, are read by export_id.

get_category_drilldown

ChatGPT
The distinct questions being asked inside one topic, deduplicated into clusters of near-identical wording. Each row has representative_text, cluster_hash, frequency (how many questionnaires asked it), member_count (how many raw questions), customers_count, knowledge_section, consistency_score/consistency_level (how consistently we answered it), and canonical_answer when one exists. Also returns total, unique_clusters and total_questions for the whole topic, all_customers, and the topic display_name. Answers questions such as 'what are the most-asked questions in <topic>' or 'which repeated questions do we answer inconsistently'; gap_only=true keeps clusters with no good answer yet. REQUIRED: knowledge_section (a custom topic UUID or 'UNCATEGORIZED', the category id in the monthly topic trends). Optional: days (30, 60, or 90) or month ('YYYY-MM'), q search, min_frequency, customer_names, sort_by (asked|customers|question), sort_order, limit/offset. Rows carry no per-customer attribution. Serves only already-computed clusters, and month is limited to a rolling window of recent months: an older month returns a 400 naming the supported range. is_warming true means the clustering for this filter combination has not been computed yet: rows are empty even when the topic has questions, and the clustering is computed in the background rather than by repeated calls.

get_category_drilldown_by_customer

ChatGPT
The clustered questions inside one topic (deduplicated near-identical wording), exploded to one row per (question cluster x customer): representative_text, customer_name, asked_by_customer (times that customer asked it), asked_total (times everyone asked it), customers_count, and knowledge_section. Also returns total, unique_clusters, total_questions, all_customers, and customer_totals per customer. Answers questions about who is asking, such as 'which customers keep asking about <topic>', 'what does Contoso ask us most', or 'is this question specific to one customer or industry-wide'. REQUIRED: knowledge_section (a custom topic UUID or 'UNCATEGORIZED', the category id in the monthly topic trends). Optional: days (30, 60, or 90) or month ('YYYY-MM'), q search, min_frequency, gap_only, customer_names filter, sort_by, sort_order, limit/offset. Serves only already-computed clusters, and month is limited to a rolling window of recent months: an older month returns a 400 naming the supported range. is_warming true means the clustering for this filter combination has not been computed yet: rows are empty even when customers asked, and the clustering is computed in the background rather than by repeated calls.

get_category_drilldown_by_questionnaire

ChatGPT
One row per questionnaire that carries a custom knowledge topic, answering 'which questionnaires were mostly about <topic>' and 'which ones had a whole section on <topic>'. Each row has questionnaire_id, questionnaire_title, customer_name, status, created_at, completed_at, total_question_count, topic_question_count, topic_share (0-1), topic_sections (each with section_display_name, topic_question_count, section_question_count), and coverage_kind: MOSTLY_TOPIC when topic questions are at least half the questionnaire, else HAS_TOPIC_SECTION when a section has at least 3 topic questions and they are at least half of that section, else SOME_TOPIC_QUESTIONS. Also returns total and coverage_kind_counts. Rows are sorted by coverage_kind then topic_question_count descending. REQUIRED: knowledge_section, which must be a custom topic UUID, the category id in the monthly topic trends ('UNCATEGORIZED' is rejected because it is not a topic). Optional: days (30, 60, or 90) or month ('YYYY-MM') filtering on when the questionnaire was uploaded, and limit/offset. Rows are questionnaires, not the questions inside the topic or the customers asking them.

get_contract_obligation_by_id

ChatGPT
The full record of one contract obligation, by obligation_id (the obligation id shown in contract obligation lists). Returns id, review_id (the signed contract), counterparty_name (the customer on that contract), obligation_text (the verbatim contract language), promise_summary, notification_target, deadline_text, frequency, due_date, category, responsible_party (VENDOR, CUSTOMER or BOTH), obligation_group_id, term_group_id, is_tightest_in_group, status, needs_review, needs_review_rationale, section_number, section_title, citation_display_text, trigger_defined_term, trigger_definition_summary and trigger_definition_citation (how the contract defines the event that starts the clock, such as its definition of Security Incident), standard_term_deviation_summary, standard_term_reference_text, renegotiate_to_standard, assigned_to_name, superseded_by_obligation_id (set when a newer extraction replaced this row), finding_id, created_at and updated_at. An unknown id returns 404. The record does not carry the source document's name or the conflicts with our public content.

get_contract_obligation_source_context

ChatGPT
Where one contract obligation comes from, by obligation_id. Returns source_file_name (the signed document the obligation was extracted from); incorporated_into, set when that document is a linked one (such as a security addendum or DPA the contract incorporates by URL) and null otherwise, with parent_review_id and parent_file_name (the contract that incorporates it), incorporation_quote (the contract sentence that pulls it in) and applicability_condition (when it applies); and public_conflicts, the open conflicts between this obligation and our public content, each with conflict_id, public_statement_text, public_source_url, conflicting_obligation_ids, explanation and status. An unknown id returns 404. It names the source document to cite and shows whether our trust center or website says something different.

get_contract_obligation_summary

ChatGPT
Headline numbers for the organization's signed-contract obligation register: total (obligations extracted from signed contracts that are not deleted), by_status (count per status, such as IDENTIFIED, ACKNOWLEDGED, IN_PROGRESS, COMPLETED, VERIFIED, WAIVED or SUPERSEDED), by_category (count per contract section category, such as SECURITY or DATA_PROTECTION), overdue_count (commitments past their due date that are not completed, verified, waived or superseded), and coverage (contract_count, customer_count, and processing_count, the signed contracts still being analyzed). Optional review_id narrows total, by_status, by_category and overdue_count to one signed contract (the id of a contract review, or the review_id of any obligation); coverage always covers the whole register. The groups field is always an empty list here, so this response holds no per-group detail, such as which customer holds the tightest breach notification term. It answers how many obligations there are per status and how many are overdue.

get_contract_obligations_by_customer

ChatGPT
Every live commitment one customer's signed contracts created, across all obligation groups: the answer to 'what exactly did we promise <customer>'. Requires customer_key, the key of a customer in the obligation register (an account id, or the counterparty name for a contract not linked to an account); a customer with no live commitments returns 404. Returns customer_key, customer_label, total_commitment_count, matching_row_count, group_count, tightest_term_count, needs_review_count and one page of commitments. Each commitment has obligation_id, review_id (the signed contract review), group_id, group_name, group_domain, promise_summary (the term in plain words, such as '72 hours after confirming a breach'), obligation_text (the verbatim contract language), notification_target (who or where we must notify), citation_display_text, section_number and section_title (where the contract says it), status, is_bespoke, is_tightest_in_group, tightest_customer_label (when this customer is not the strictest in the group, the customer who is), needs_review and needs_review_rationale. Rows come tightest term first, then by group name. Params: q (2 or more characters) searches the group name, the promised term, the customer who sets the tightest term and the citation, so q='breach' narrows to breach notification commitments; review_state_filter (needs_review, clear), term_type_filter (bespoke, standard), status_filter (IDENTIFIED, ACKNOWLEDGED, IN_PROGRESS, COMPLETED, VERIFIED, WAIVED) and domain_filter (SECURITY, PRIVACY, LEGAL), each comma-separated; sort_by (obligation, promise, tightest, citation) with sort_order asc or desc; limit 1-100 (default 25) and offset.

get_contract_obligations_by_group

ChatGPT
Every customer's commitment inside one obligation group, the answer to 'which customers have a 24, 48 or 72 hour breach notification commitment and what exactly did we promise each one'. Requires group_id, the id of an obligation group in the obligation register; an unknown group_id returns 404. Returns group_id, group_name, plain_question, domain, total_obligations, total_reviews_with_group, total_reviews_analyzed, customer_count, has_notification_targets, rendered_row_count, matching_row_count and one page of obligations. Each obligations row has obligation_id, review_id (the signed contract), account_id, counterparty_name and counterparty_names (rows that carry the same term are merged into one row naming every customer on it, and member_obligations lists each customer's own obligation_id), promise_summary (the term in plain words, such as '24 hours, even for suspected incidents'), obligation_text (the verbatim contract language), notification_target (who or where we must notify), deadline_text, frequency, citation_display_text, section_number and section_title (where the contract says it), is_tightest_in_group, is_bespoke, status, merged_statuses, needs_review, needs_review_rationale, trigger_defined_term and trigger_definition_summary (how the contract defines the event that starts the clock, such as its definition of Security Incident), standard_term_deviation_summary and standard_term_reference_text (how the term differs from our standard paper). Rows come tightest term first. The response also has standard_terms_group (customer_names, customer_count and shared_term_text for customers on our standard term) and public_conflicts (open conflicts between these commitments and our public content); both cover the whole group and are never paged or filtered. Params: q (2 or more characters) searches customer names, the promised term and the citation; review_state_filter (needs_review, clear), term_type_filter (bespoke, standard), status_filter (IDENTIFIED, ACKNOWLEDGED, IN_PROGRESS, COMPLETED, VERIFIED, WAIVED) and domain_filter (SECURITY, PRIVACY, LEGAL), each comma-separated; sort_by (customer, promise, notification_target, citation) with sort_order asc or desc; limit 1-100 (default 25) and offset.

get_contract_review_analytics

ChatGPT
Legal (contract) review analytics: total_reviews, completed_reviews, total_findings, finding_status_breakdown (meet, favorable, fail_to_meet, unsure, not_applicable), avg_turnaround_hours, avg_findings_per_review and daily_activity (reviews and findings per day). Covers aggregate contract-review throughput and finding quality; it does not list the individual contract reviews behind these numbers. Params: time_frame (month/quarter/halfyear/year/alltime/custom) or start_date/end_date in microseconds.

get_contributor_participation

ChatGPT
Per-contributor AND per-team questionnaire activity. The per-user breakdown is inside this response; there is no deeper per-user endpoint. contributors[] carries user_id, name, email, questionnaires_worked_on, questions_approved, questions_edited, total_questions, questions_assigned/completed, questionnaires_assigned/completed, comments_count, facts_added, uploaded_questions/questionnaires and avg_turnaround_hours; teams[] carries the same counts aggregated per group with group_name and member_count. Answers 'who is doing the work', workload balance, and team contribution comparisons. Params: time_frame or start_date/end_date in microseconds, limit (max contributors, default 200).

get_conversation

ChatGPT
Get full conversation history by ID. Returns all messages in a conversation including AI responses with their citations.

get_conversation_analytics

ChatGPT
Aggregate chat activity: total_conversations, unique_users, total_questionnaires, total_questions, average_questions_per_questionnaire, the conversation-volume time_series_data, weekly_active_users (week + unique_users), and per-user stats (name, email, conversation_count, last_active_at, primary_source). Answers 'how many chats', 'who is using chat', and weekly-active-user trends. It carries counts only, not the dollar or hours-saved value of chat activity and not the individual conversations. Params: time_frame (month/quarter/halfyear/year/alltime/custom) or explicit start_date/end_date in microseconds.

get_conversation_insight

ChatGPT
Full detail for ONE chat conversation by conversation_id (the id of a row in the chat conversation log): question, answer, asking user, source_type, answered_at, plus feedback_positive/feedback_comment, confidence_score and confidence_reasoning, citations, follow_ups, query_explanation, query_answer_requirements, query_peer_answer, and the agent_trace. It is the only chat analytics view that carries citations and the agent trace.

get_document_content

ChatGPT
Read the full raw content of a document by its ID. Returns the complete text (Notion page as markdown, Slack conversation, uploaded file content) along with metadata, extracted facts, and the original source URL. document_id is the source document ID that knowledge search results and the knowledge source list return.

get_document_download_urls

ChatGPT
Get download URLs for documents by their IDs. Returns short stable download links for file-backed sources (Google Drive, uploaded files, etc.) that are valid for 24 hours, direct URLs for web-backed sources (Notion pages, Confluence, websites), or null for sources that don't have downloadable files (Slack). IDs that do not match an approved, completed document are listed in not_found_ids and get no row in results. A download link stops working if any part of it changes. document_ids are source document IDs, as returned by knowledge search results, the knowledge source list, and document reads.

get_editorial_guide

ChatGPT
Get your organization's editorial guide: the writing-style and tone rules the answering agent follows when drafting questionnaire answers, which shows the current guidance and how answers are styled.

get_full_documents_from_knowledge_base

ChatGPT
Retrieve full underlying KB documents (raw content + facts + source URL) ranked by relevance OR by ID — closes the search→fetch round-trip into one call. Returns the document bodies themselves, without going through the chat agent. Two mutually-exclusive modes: - Query mode: pass query to get top-K full documents semantically ranked by similarity. - By-ID mode: pass document_ids to batch-fetch full content for known IDs. Parameters: - query (string, optional): natural language search text. Provide this OR document_ids. - document_ids (array of strings, optional): UUIDs to fetch. Provide this OR query. - source_types (array of strings, optional, query mode only): filter by source type. Valid values: BOX_INTEGRATION, CLARI_COPILOT_INTEGRATION, CONFLUENCE_INTEGRATION, CORE_KNOWLEDGE, CORRECTIONS, DRATA_INTEGRATION, EMAIL_INTEGRATION, GITHUB_INTEGRATION, GITLAB_INTEGRATION, GLEAN_INTEGRATION, GONG_INTEGRATION, GOOGLE_DRIVE_INTEGRATION, GURU_INTEGRATION, HUBSPOT_INTEGRATION, KNOWLEDGE_ANSWER, LINEAR_INTEGRATION, MANUAL_FACT, MINTLIFY_INTEGRATION, MOMENTUM_IO_INTEGRATION, NOTION_INTEGRATION, ONEDRIVE_INTEGRATION, ONELEET_INTEGRATION, PRE_APPROVED_ANSWER, QUESTIONNAIRE_KNOWLEDGE, RIPPLING_INTEGRATION, SALESFORCE_INTEGRATION, SHAREPOINT_INTEGRATION, SLAB_INTEGRATION, SLACK_INTEGRATION, TRUST_CENTER, VANTA_INTEGRATION, WEBSITE_INTEGRATION. - business_segments (array of strings, optional, query mode only): filter by organization-defined business segment names. - limit (integer, optional, default 5, clamped to 1..20): maximum number of documents to return in query mode. - max_content_length_per_doc (integer, optional, default null = use server cap of 100000 chars): truncate each document body to this many characters; flagged via content_truncated. Server hard cap of 100000 chars always applies. - include_facts (boolean, optional, default true): whether to include extracted facts per document. Hard limits: query mode limit is clamped to 1..20; per-document content is clamped to at most 100000 characters; by_id mode returns an error for a call with more than 25 document_ids.

get_individual_questions

ChatGPT
Individual question-level rows, the drill-down beneath the per-questionnaire edit rollup. Each row has question_text, question_number, sheet_name, its questionnaire_id/questionnaire_title/company_name, status, confidence_score, is_approved, approved_by_name, edited_by_names and edited_by_count, edit_classification, times_asked_count, answer_reasoning, source_names, and response_diffs pairing the AI answer against the human answer field by field. Also returns total, edited_count, blank_filled_count, unanswered_count, and edit_classification_counts. edited_count means a human CHANGED a real AI-generated answer (factual or style edit); blank_filled_count means the AI left the question blank and a human filled it in (NOT an edit); answers awaiting async classification are in edit_classification_counts.pending_classification and are excluded from both. The question_filter=edited row filter still returns every human-touched question, including blank fills. Shows the actual questions a human rewrote or left blank, and exactly what was changed. questionnaire_id scopes the rows to one questionnaire and question_filter=edited|unanswered isolates a problem set; sort_by/sort_order and limit/offset order and page the rows. Window: time_frame, or explicit start_date/end_date in microseconds.

get_mcp_action_approval

ChatGPT
Read the outcome of a change that waits for a person's approval in Wolfia (read only). Takes the approval_id a tool returned with status approval_required_in_wolfia. Returns status (PENDING, APPROVED, DECLINED, EXPIRED, FAILED, EXECUTED), when it was decided, the result summary and result once it ran, or the error. It never approves anything; only a person can, on the approval_url page.

get_pre_approved_answer

ChatGPT
Full detail of one pre-approved answer, by pre_approved_id (the id of a pre-approved answer). Returns id, question, answer (the company-wide default answer, null when the question only has variants), original_question and original_answer (the text before anyone edited it), variants (answers for specific business segments, each with id, answer, internal_note, original_answer and tags, the business segment ids it applies to), default_answer_tag_ids (the business segments the default answer covers), is_active, is_approved, approved_at and approver, category_ids, citation_count (how many times answering used it), expires_at, freshness_tracking_enabled, internal_note, creator, updater, and created_at and updated_at in microseconds. Returns a 404 error when the id does not exist or the answer was deleted.

get_pre_approved_answer_performance

ChatGPT
Monthly pre-approved-answer (PAA) performance for an organization: library growth (paa_created, paa_cumulative_active), the match rate (pa_match / ai_answered — the share of AI-drafted answers that matched a pre-approved answer instead of being AI-generated), and the human edit rate on matched pre-approved answers (pa_match_all_edit / pa_match and pa_match_factual_edit / pa_match — how often a person changed a matched pre-approved answer, and the subset that changed factual content). Months are keyed by questionnaire created month for the match/edit series and by pre-approved-answer created month for library growth. Window: time_frame, or explicit start_date/end_date in microseconds.

get_questionnaire_answers

ChatGPT
List one questionnaire's questions with their answers, one page at a time. Takes questionnaire_id; answer_filter, one of all, needs_attention, ready_for_review, unanswered, approved, not_approved; question_ids to restrict the result to specific questions; and offset and limit for paging, where limit is at most 100. Each question carries question_id, its 1-based position, section name, question text, status, confidence score, review recommendation and reasons, reviewer_note (why the answer may need a person), answer_reasoning, approval and out-of-scope flags, needs_attention, is_unanswered (every field empty), and its fields, each with response_id, field name, field type, required flag, options for select fields, and the answer. A field's response_id identifies that field when its answer is edited; question_id identifies the question when it is approved or regenerated. Returns total_questions, total_matching, offset, limit and next_offset, which is null on the last page.

get_questionnaire_approval_history

ChatGPT
Get approval history for a single questionnaire by id: when it was received, its assignee, total vs approved questions, the approval time-span, per-approver contribution, an activity window, and per-question approval events (timestamps in microseconds since epoch). Use this to inspect the level-of-effort for one questionnaire.

get_questionnaire_edit_summaries

ChatGPT
Per-questionnaire rollup of what humans did to the AI's answers: one row per completed questionnaire with questionnaire_title, company_name, completed_at, total_questions, edited_count, unanswered_count, accepted_as_is_count, user_verified_count, and edit_classification_counts (minor_factual, minor_style, major_factual, major_style, no_edit, filled_empty_field, pending_classification), plus a total row count and filter-wide aggregates (edited_count_across_all_questionnaires, edit_classification_counts_across_all_questionnaires) computed over every matching questionnaire regardless of limit/offset. edited_count means a human CHANGED a real AI-generated answer (factual or style edit). Blank fills (the AI left the question blank and a human filled it in) are NOT edits; they appear as filled_empty_field in edit_classification_counts and via the blank_filled answer status. Answers awaiting async classification appear as pending_classification and are excluded from edited_count. Shows which questionnaires were edited most and what kind of editing it was. This is the rollup level: rows carry counts, not the individual questions or the AI-versus-human answer diffs, and not organization-wide ROI totals. Supports search, answer_statuses (comma-separated: accepted_as_is, user_verified, edited, blank_filled, unanswered), sort_by/sort_order, and limit/offset. Window: time_frame, or explicit start_date/end_date in microseconds.

get_questionnaire_export

ChatGPT
Read one questionnaire export's status by export_id. Takes questionnaire_id and export_id, the ids returned when the export was started. Returns export_status, completed_questions, total_questions, and, once the export is complete, file_name, mime_type, file_size_bytes, and file_content_base64: the exported file itself, base64-encoded, which decodes to the file named file_name. download_url is a Wolfia link to the same file, valid for 24 hours, that opens for a person signed in to this Wolfia workspace.

get_questionnaire_instruction_version

ChatGPT
Get a specific version of your organization's questionnaire answering instructions by version number (full instruction text plus metadata). version_number is a number from the instruction version history; an unknown version returns 404. Requires admin or expert role.

get_questionnaire_performance

ChatGPT
Per-questionnaire operational scorecard: one row per questionnaire with title, company_name, status, created_at/updated_at, total_questions, ai_answered, manual_answered, acceptance_rate, fill_rate, avg_confidence, turnaround_hours, and assignee_names, plus a total row count. Ranks and finds questionnaires ('which questionnaires had the worst acceptance rate', 'what did we send Contoso', 'what is assigned to this teammate') with sort_by/sort_order, status_filter, assignee_id, company_name_filter, a q search, and limit/offset. It has no organization-wide totals or ROI, no per-questionnaire edit classification breakdown, and no AI-versus-human time split. Window: time_frame, or explicit start_date/end_date in microseconds.

get_questionnaire_revenue_metrics

ChatGPT
QUESTIONNAIRE-attributed revenue: CRM pipeline for opportunities that have a security questionnaire linked to them in Wolfia. Returns metrics (won_revenue, won_deal_count, pipeline_revenue and pipeline_deal_count for still-open deals, total_questionnaire_count) plus top_won_deals and top_pipeline_deals, each deal row carrying account_name, opportunity_name, domain, amount, stage, and questionnaire_count (how many questionnaires Wolfia answered for that opportunity). Top lists are the 10 largest by amount. Answers questions such as 'how much revenue did answering questionnaires unlock' or 'which deals needed the most questionnaire work'. Open opportunities whose CRM amount is implausible are left out of pipeline_revenue and top_pipeline_deals: an amount whose magnitude exceeds ten times this tenant's largest won-deal amount. A negative amount is a normal contraction (debooking, churn, credit memo, amendment) and is counted like any other as long as its magnitude is within that cap. won_revenue is never filtered. excluded_open_opportunities reports that filtering with excluded_count, excluded_amount, and up to five of the excluded deals (crm_opportunity_id, account_name, opportunity_name, amount), largest magnitude first. A non-zero excluded_count means the tenant's CRM holds implausible amounts, so pipeline_revenue omits those deals and is not the whole open pipeline. The excluded deals are still counted in pipeline_deal_count. This is the questionnaire attribution model. Trust-center-attributed revenue, matched through trust center visits on account email domains, is a separate attribution model: a deal can appear in one, both, or neither, so the two totals overlap and do not add up to a combined figure. Requires a connected CRM (Salesforce or HubSpot); returns all-zero metrics and empty deal lists when no CRM is connected. Takes no date filters: it covers all linked opportunities.

get_questionnaire_status

ChatGPT
Read one questionnaire's status and question counts: title, company name, status code and display name, whether answering is finished or failed, whether it is waiting for a reviewer to start it, per-status question counts, where answered_questions counts every question that finished processing, including declined, skipped and manually cleared ones (the per-question is_unanswered flag marks the empty ones), whether the caller can edit and approve answers, the Wolfia web page URL, and timing: uploaded_at, extraction_completed_at and answering_completed_at (UTC), extraction_seconds, answering_seconds, total_seconds_to_answers, and seconds_since_upload. Takes questionnaire_id, the id returned when a questionnaire is submitted and listed with each questionnaire in the workspace. The questions and their answers are not included.

get_questionnaire_turnaround

ChatGPT
How long questionnaires take, split into machine time and human time. Returns a summary (avg_ai_time_hours, avg_human_time_hours, avg_total_turnaround_hours, questionnaire_count), one row per completed questionnaire (title, company_name, status, total_questions, ai_time_hours, human_time_hours, total_turnaround_hours, uploaded_by_name), and a trend series of median_ai_hours vs median_human_hours per day. Answers 'how fast do we turn questionnaires around', 'where is the time going, AI or review', or 'is turnaround improving'. It measures time only, not accept/edit quality. Window: time_frame, or explicit start_date/end_date in microseconds.

get_questionnaire_value_analytics

ChatGPT
Organization-wide questionnaire answering scorecard for a time window. It covers three areas: (1) AI answer performance: acceptance_stats with total_ai_answers, accepted_as_is, edited_by_user, user_verified, human_edited_ai (a human changed a real AI-generated answer, a factual or style edit), human_filled_gap (the AI left the question blank and a human filled it in; NOT counted as an edit), questions_left_blank, acceptance_rate, fill_rate (human-modified answers still awaiting async edit classification are excluded from human_edited_ai, human_filled_gap, and acceptance_rate, so edited_by_user can exceed the sum of those buckets); (2) questionnaire volume: total_questionnaires, total_questions_answered, unique_approvers, time_period_days, and per-period time_series_data bucketed at time_series_interval (day for windows up to 31 days, week up to 180, month beyond; time_series_interval is the unit of every bucket); (3) hours saved and ROI: time_savings (total_hours_saved, accepted_hours, editing_hours, per_question_avg_minutes) and value_range with conservative (min) and optimistic (max) dollar totals split into accepted_answer_value and edited_answer_value, plus the analyst-rate methodology behind them. Answers 'how much time or money did Wolfia save us', 'what is our AI acceptance/edit rate', and 'how many questionnaires/questions did we do'. It returns organization-wide totals only, with no per-questionnaire breakdown. The dollar figures use Wolfia's product-default analyst rates and do NOT reflect custom ROI assumptions the organization has saved via /insights/impact-assumptions, so they are a default-rate estimate and can differ from the Analytics Impact dashboard. Window: time_frame, or explicit start_date/end_date in microseconds.

get_review_contract

ChatGPT
One contract review in full, by review_id (the id of a contract review, also carried as review_id on every contract obligation). The caller sees a review only when it is shared with the organization or the caller uploaded it; another member's private review returns 404, the same as an unknown id. Signed contracts are always shared with the organization. Returns id, file_name, status, review_stage, is_signed_contract, is_org_document, detected_role and confirmed_role (which side of the contract we are), obligation_counterparty_name, detected_parties (party_name, party_role), section_count, obligation_count, obligations_extracted_at, error_message, created_by_name, created_at and updated_at; benchmark (contract_type, overall_position ABOVE_MARKET, AT_MARKET or BELOW_MARKET, position_summary, missing_clauses); and sections, each with section_number, section_title, section_text, category, applies_to and findings. Each finding has id, status (MEET, FAVORABLE, FAIL_TO_MEET, UNSURE, NOT_APPLICABLE), reasoning, target_text (the contract language it is about), suggested_response, topic, defined_term, referenced_sections, kb_evidence (fact_id, statement and source_name of the knowledge-base facts it was checked against), resolution, adopted_text, resolution_comment, resolved_by_name, benchmark_reasoning, market_position and industry_standard_value. The response carries the whole contract text, so its size follows the length of the document, and it covers one review per call. It does not list the commitments a signed contract created; those are in the obligation register.

get_slack_auto_respond_instructions

ChatGPT
Get your organization's Slack auto-respond instructions: the org-level guidance the agent follows when auto-answering questions in Slack, plus the insufficient-knowledge fallback message and respond-once-per-thread setting, which together describe the current Slack answering behavior.

get_slack_thread

ChatGPT
Read a complete Slack thread with all replies. Takes the channel name and the thread timestamp (thread_ts, as carried by Slack message search results) and returns the full conversation with sender names and message content.

get_token_usage

ChatGPT
Estimated LLM token consumption for the organization: total_tokens, product_count, product_breakdown (conversations, knowledge_base, contract_review, questionnaire, trust_center) and daily_product_usage (the same per-product split per day for trend charts). Answers 'how much are we using Wolfia across products' and usage-trend questions. Params: time_frame or start_date/end_date in microseconds.

get_top_open_deals_by_account

ChatGPT
Ranked list of accounts by OPEN pipeline revenue that the trust center influenced: accounts that visited the trust center, joined to their still-open CRM opportunities by matching the account's primary domain plus every additional domain linked to that account (an admin-unlinked secondary domain is excluded; an account's own primary domain keeps counting even if its own link row was unlinked, and it stops counting only when a different account actively links it, in which case that other account is credited instead) against the CRM opportunity domain. Opportunities the CRM marks deleted are excluded. Each row has name (the account name from that account's top-ranked non-deleted won-or-open CRM opportunity, ranked by amount with unpriced opportunities last, then by how many opportunities carry the same account name, then by the most recently created, falling back to the stored trust-center account name when the CRM has none; the same label the won-revenue account ranking shows for the same account, and it does not shift with the requested window), domain (the account's primary domain), account_id, logo_url, revenue_associated (summed amount of that account's open deals across all of its linked domains), and days_open (age of that account's oldest open deal across those domains). Paginated with total_count/page/page_size/total_pages. Answers questions such as 'which live deals is the trust center touching', and days_open shows stalled pipeline. Open opportunities whose CRM amount is implausible are left out of revenue_associated (an amount whose magnitude exceeds ten times this tenant's largest won-deal amount); ordinary negative contractions are counted, and the trust-center revenue totals list the excluded deals in excluded_open_opportunities. It covers open pipeline only, not won revenue. revenue_associated sums to influenced.pipeline_revenue in the trust-center revenue totals for the same window, except that accounts with no open pipeline are omitted. It is trust-center attributed, not questionnaire attributed: questionnaire-attributed pipeline is a separate attribution model. Requires a connected CRM (Salesforce or HubSpot); returns an empty list when no CRM is connected. start_date/end_date (microseconds) filter on opportunity creation date.

get_top_won_deals_by_account

ChatGPT
Ranked list of accounts by CLOSED-WON revenue that the trust center influenced: accounts that visited the trust center, joined to their CRM opportunities by matching the account's primary domain plus every additional domain linked to that account (an admin-unlinked secondary domain is excluded; an account's own primary domain keeps counting even if its own link row was unlinked, and it stops counting only when a different account actively links it, in which case that other account is credited instead) against the CRM opportunity domain. Opportunities the CRM marks deleted are excluded. Each row has name (the account name from that account's top-ranked non-deleted won-or-open CRM opportunity, ranked by amount with unpriced opportunities last, then by how many opportunities carry the same account name, then by the most recently created, falling back to the stored trust-center account name when the CRM has none; the same label the open-pipeline account ranking shows for the same account, and it does not shift with the requested window), domain (the account's primary domain), account_id, logo_url, and revenue_associated (summed amount of that account's won deals across all of its linked domains). Paginated with total_count/page/page_size/total_pages. Answers questions such as 'which customers closed after using our trust center' or 'top accounts by won revenue'. It covers won revenue only, not still-open pipeline. revenue_associated sums to the influenced.won_revenue figure in the trust-center revenue totals for the same window (identical account set, domain matching, and deleted-opportunity exclusion), except that accounts with no won revenue in the window are omitted, so the row count is not influenced.won_deals. It is trust-center attributed, not questionnaire attributed: questionnaire-attributed deals are a separate attribution model. Requires a connected CRM (Salesforce or HubSpot); returns an empty list when no CRM is connected. start_date/end_date (microseconds) filter on deal close date.

get_trust_center_revenue_metrics

ChatGPT
TRUST-CENTER-attributed revenue: CRM pipeline for accounts that visited the trust center, attributed by matching each account's primary domain plus every additional domain linked to that account (an admin-unlinked secondary domain is excluded; an account's own primary domain keeps counting even if its own link row was unlinked, and it stops counting only when a different account actively links it, in which case that other account is credited instead) against the CRM opportunity domain. Opportunities the CRM marks deleted are excluded from both blocks. Returns two side-by-side blocks, influenced (trust-center-touched accounts only) and total (all CRM opportunities), each with won_revenue, pipeline_revenue (open deals), won_deals, open_deals, total_opportunities, avg_deal_size, and avg_days_to_close, plus time_series_data (per-period won/pipeline revenue and deal counts, auto-bucketed day/week/month from the requested window). Open opportunities whose CRM amount is implausible are left out of every pipeline sum here: an amount whose magnitude exceeds ten times this tenant's largest won-deal amount. A negative amount is a normal contraction (debooking, churn, credit memo, amendment) and is counted like any other as long as its magnitude is within that cap. won_revenue is never filtered. Each block's excluded_open_opportunities reports that filtering with excluded_count, excluded_amount, and up to five of the excluded deals (crm_opportunity_id, account_name, opportunity_name, amount), largest magnitude first. A non-zero excluded_count means the tenant's CRM holds implausible amounts, so pipeline_revenue omits those deals and is not the whole open pipeline. The excluded deals are still counted in open_deals and total_opportunities. Answers questions such as 'how much revenue did the trust center influence' and compares influenced revenue against the whole book. This is the trust center attribution model. Questionnaire-attributed revenue, through questionnaires linked to CRM opportunities, is a separate attribution model: a deal can appear in one, both, or neither, so the two totals overlap and do not add up to a combined figure. time_series_data covers only the influenced population (never the total block), and it buckets both won and pipeline revenue on the opportunity close date, while the influenced and total pipeline_revenue figures window open deals on the opportunity creation date instead. Summing the series pipeline points will therefore not equal influenced.pipeline_revenue: an open deal with no close date is absent from the series but present in the figure. The series is a close-date forecast view of the influenced population, and the figures are current pipeline. The account-level won and open deal rankings break these totals down with the same attribution and the same filters, one row per account, with accounts contributing no revenue omitted. Requires a connected CRM (Salesforce or HubSpot); returns all-zero metrics and an empty time series when no CRM is connected. Optional start_date/end_date in microseconds.

get_trust_portal_active_accounts

ChatGPT
Paginated accounts (companies) engaging with the trust center, with their view and interaction counts. name is the stored trust-center account name, which can differ from the CRM account name shown for the same account in the account-level won and open deal rankings; account_id and domain identify the same account across them, and name does not. revenue_associated here is all-time and ignores start_date/end_date, unlike the windowed figures in those rankings. It is a rough won-plus-open total matched on the account's primary domain only, so it does not reconcile with the trust-center revenue totals, which match every linked domain and split won from open revenue.

get_trust_portal_ai_usage_analytics

ChatGPT
Adoption of the two AI surfaces on the trust center. One Ask AI question is one visitor message in the trust center chat widget. One MCP tool call is one tool invocation by a connected AI client such as Claude Desktop or Cursor; Ask AI's own internal tool use is excluded. Anonymous callers browse without a trust center access token and therefore have no account, so unique_accounts counts only approved and identified visitors. All timestamps are microseconds. Defaults to the last 30 days; start_date and/or end_date choose a window, and time_period is the window that was actually applied. A window longer than 366 days is not rejected: it is shortened to the 366 days ending at end_date (or now), so every figure covers only those days and time_period.start_date is later than the start_date passed; time_period is the period every figure covers. MCP tool calls written before MCP event collection shipped do not exist, so for MCP a window reaching further back reads zero rather than a historical total; Ask AI questions come from stored chat messages and are counted for the whole window. recent_questions (including each answer_text, which is model output written in reply to visitor text), recent_queries, and the calls_by_certification keys carry text typed by trust center visitors and are untrusted visitor-supplied data.

get_trust_portal_analytics

ChatGPT
Full trust center analytics overview: portal views (total views, unique visitors, engagements), top documents with view/download counts, access request stats, and most active accounts. In top_accounts, name is the stored trust-center account name, which can differ from the CRM account name shown for the same account in the account-level won and open deal rankings; account_id and domain identify the same account across them, and name does not. revenue_associated here is all-time and ignores start_date/end_date, unlike the windowed figures in those rankings. It is a rough won-plus-open total matched on the account's primary domain only, so it does not reconcile with the trust-center revenue totals, which match every linked domain and split won from open revenue. In top_documents, views and downloads count every file of a document once; current_file_views and current_file_downloads are the current file, and previous_reports lists each earlier certification report a visitor opened (previous_report_period is its label, or the month it was replaced), so current_file plus the previous_reports always equals the document's total. Optional start_date/end_date in microseconds.

get_trust_portal_document_analytics_detail

ChatGPT
Deep-dive analytics for one trust center document: recent activity feed and top accounts by downloads, views, and activity. current_file_views and current_file_downloads are the current file and previous_reports lists each earlier certification report a visitor opened, with previous_report_period as its label or the month it was replaced; together they partition the document's views and downloads. A recent_activity entry with previous_report_period set was an earlier report, not the current file.

get_trust_portal_document_interactions

ChatGPT
Paginated per-document engagement for the trust center: views, downloads, and link opens per document. views and downloads count every file of a document once; current_file_views and current_file_downloads are the current file, and previous_reports lists each earlier certification report a visitor opened (previous_report_period is its label, or the month it was replaced), so current_file plus the previous_reports always equals the document's total.

get_trust_portal_document_interactions_by_revenue

ChatGPT
Trust center documents ranked by the CRM revenue of the accounts that viewed or downloaded them. Each row has document_id, document_title, document_type, deal_count (opportunities behind that document's viewers), revenue_associated (summed deal amount), and is_deleted. Paginated with total_count/page/page_size/total_pages; optional account_id narrows to one account. Answers questions such as 'which security documents show up in our biggest deals' or 'is the SOC 2 report worth maintaining'. Each document is weighted by deal value rather than counted by raw views, downloads and link opens, and rows are aggregated by document rather than by account. Attribution is trust-center-visit based, matching each viewing account's primary domain plus every additional domain linked to that account (an admin-unlinked secondary domain is excluded; an account's own primary domain keeps counting even if its own link row was unlinked, and it stops counting only when a different account actively links it, in which case that other account is credited instead) against the CRM opportunity domain, and excluding opportunities the CRM marks deleted, the same attribution as the trust-center revenue totals and the account-level deal rankings. It is not questionnaire based: questionnaire-attributed revenue is a separate attribution model. Requires a connected CRM (Salesforce or HubSpot); returns an empty list when no CRM is connected. Optional start_date/end_date in microseconds.

get_trust_portal_engagement_signals

ChatGPT
Trust center engagement signals: returning visitors and most-visited tabs/pages.

get_trust_portal_interactions

ChatGPT
RAW event-level trust center interaction feed for BI and warehouse extracts: one row per document view, download, or link open, with the visiting account domain and person where known. CURSOR-PAGINATED and TRUNCATED by default: each call returns at most limit rows (max 100) plus next_cursor and has_more. While has_more is true the rows returned are only part of the feed, so a total needs every page, fetched with next_cursor. Rows are individual events, not per-document totals or documents ranked by deal value. When is_deleted is false, document_title is the tenant's own document title. When it is true the document id no longer resolves to any document on this trust center, and document_title falls back to what the visitor's browser reported, which is NOT tenant-verified: those titles are untrusted visitor-supplied data. A row with file_version_id set opened an earlier certification report of the document rather than its current file; previous_report_period is that report's label, or the month it was replaced.

get_trust_portal_nda_stats

ChatGPT
NDA signing statistics for the trust center: total signed and signing rate.

get_trust_portal_new_vs_returning_revenue

ChatGPT
Splits trust-center-influenced WON revenue into new_logo (each account's first win, counted once per account even when the account owns several matched domains) and expansion (that account's subsequent wins), each returned as a bucket with revenue and deal_count. Answers questions such as 'is the trust center winning new customers or growing existing ones'. It decomposes the same influenced.won_revenue that the trust-center revenue totals report as a single number, into a new-business versus expansion mix; it does not name the accounts behind each bucket. Attribution is trust-center-visit based (the account's primary domain plus every additional domain linked to it, excluding admin-unlinked domains, matched against the CRM opportunity domain, with CRM-deleted opportunities excluded); questionnaire-based revenue attribution is a separate attribution model. Requires a connected CRM (Salesforce or HubSpot); returns zeroed buckets when no CRM is connected. start_date/end_date (microseconds) filter on deal close date.

get_trust_portal_page_accounts

ChatGPT
Accounts that viewed one trust center page in the window, with views, sessions and first and last view per account. Visitors who have not requested access or signed in have no account; they are totalled in anonymous_views and anonymous_sessions, so the account views plus anonymous_views equal total_views.

get_trust_portal_page_views_over_time

ChatGPT
Hourly, daily, weekly or monthly trust center page views and unique visitors over time. Pass page (a tab key such as updates or documentation) to count only views of that page: tab opens plus page loads that landed on it.

get_trust_portal_pages_engagement

ChatGPT
Views per trust center page (overview, documentation, updates, controls, subprocessors and custom pages). A view is a tab open or a page load that landed on the page. accounts counts distinct accounts; anonymous visitors are not accounts. Built-in pages are always listed, with zeros when nobody viewed them in the window.

get_trust_portal_settings

ChatGPT
Get your organization's current Trust Center (trust portal) settings: access-request rules, NDA/watermark config, domain blocking, and customer-upload options for the public trust portal. Requires admin role.

get_trust_portal_subscriber_stats

ChatGPT
Trust center update subscriber statistics: active, new, and unsubscribed counts.

get_value_analytics

ChatGPT
Chat-side ROI: estimated hours saved and dollar value, computed from Wolfia's product-default rate assumptions. These figures do NOT reflect any custom ROI assumptions the organization has saved via /insights/impact-assumptions, so they are Wolfia's default-rate estimate, not the organization's own numbers, and can differ from the dollar values shown on the Analytics Impact dashboard. Returns value_range (min/max estimated dollar value split into sme_cost and asker_cost), detailed_sme_cost and detailed_asker_cost (hours saved and dollar value broken out into context-switch, answering, search and wait time), sessions (total_sessions, total_questions, avg_questions_per_session), questions_by_source, unique_users, session_breakdown (single- vs multi-question sessions), off_hours distribution, time_period_days, and the full methodology behind every figure. It covers chat only: no conversation-volume trend, no weekly active users, and no questionnaire-side ROI. Params: time_frame (month/quarter/halfyear/year/alltime/custom) or start_date/end_date in microseconds.

list_conflicts_with_knowledge_base

ChatGPT
The Conflicts tab of the Knowledge updates page: open cards from the last 90 days where facts in the knowledge base disagree with each other or with a reviewer's answer. Returns questionnaire_cards and chat_cards, each newest first, questionnaire cards before chat cards across pages. total is how many cards are open and next_offset is the offset of the next page, null on the last page. limit defaults to 20 and can be at most 20. Closing a card moves later cards to lower offsets, so offsets read before a card was closed no longer point at the same cards. Each card has verdict_id, verdict_category, card_type, missing_knowledge_summary, drafted_fact_title and drafted_fact_text (a suggested fact, when one was drafted), conflicting_items (each with item_id, item_kind, item_text, source_name and business_segment_names), old_claim_quote, new_claim_quote, is_waiting_for_approval, account_name for a card about one customer account and created_at in microseconds; a questionnaire card also has questionnaire_id, questionnaire_title, question_id and question_text, and a chat card has conversation_id, message_id and message_snippet. A conflict is settled by its verdict_id together with the item_id values of the conflicting_items whose item_kind is FACT that should stop being used, plus a title and fact_text when the card has a drafted_fact_text; a card that needs no change closes when it is dismissed.

list_contract_obligation_customers

ChatGPT
One page of the signed-contract obligation register seen by customer: one row per counterparty across every live commitment in the executed contracts, showing which customers carry the most or the tightest commitments. Each row in customers has customer_key (the key that identifies this customer when reading its commitments; it is the account id when the contract is linked to an account, otherwise the counterparty name), customer_label (the display name), contract_count, commitment_count, group_count, tightest_term_count (in how many obligation groups this customer holds us to the strictest term), needs_review_count and conflict_count (open public-content conflicts plus unresolved high-severity questionnaire parity findings). Also returns matching_count, total_count and coverage (contract_count, customer_count, processing_count). Params: q (2 or more characters) searches the customer label; review_state_filter (needs_review, clear), conflict_filter (has_conflicts, no_conflicts) and domain_filter (SECURITY, PRIVACY, LEGAL), each comma-separated; sort_by (customer, contract_count, commitment_count, tightest_term_count, needs_review_count, conflict_count), default most commitments first, with sort_order asc or desc; limit 1-100 (default 25) and offset.

list_contract_obligation_groups

ChatGPT
One page of the signed-contract obligation register grouped by kind of commitment (for example breach or incident notification, audit rights, data deletion, log retention), covering one kind of commitment across every customer, such as 'which customers have a 24, 48 or 72 hour breach notification commitment'. Each row in groups has group_id, group_name, plain_question (the question the group answers, such as 'How fast must we tell them about a breach?'), domain (SECURITY, PRIVACY or LEGAL), commitment_count, customer_count, bespoke_customer_count, needs_review_count, conflict_count, tightest_obligation_id, tightest_promise_summary (the strictest term any customer holds us to in that group, such as '24 hours'), tightest_counterparty_name (the customer who set it) and customer_labels (every customer with a commitment in the group). Also returns matching_count, total_count and coverage (contract_count, customer_count, processing_count). Only groups with at least one live commitment from a signed contract are listed. Params: q (2 or more characters) searches the group name, plain_question, the tightest term and the customer who set it, so q='breach' or q='notification' finds the breach notification group; review_state_filter (needs_review, clear), conflict_filter (has_conflicts, no_conflicts), term_type_filter (bespoke, standard) and domain_filter (SECURITY, PRIVACY, LEGAL), each comma-separated; sort_by (group_name, tightest_term, set_by, customer_count, needs_review_count), default seeded groups first, with sort_order asc or desc; limit 1-100 (default 25) and offset. customer_labels grows with the number of customers in a group, so a page's size grows with the contract book. group_id identifies the group whose per-customer terms are listed one row per customer in the obligations of that group.

list_contract_questionnaire_parity_findings

ChatGPT
Paginated findings where a signed-contract obligation contradicts a questionnaire answer we gave the same account, for example a contract that promises 24 hour breach notice while a security questionnaire answered 72 hours. Each row in findings has id, account_id, account_name, obligation_id, obligation_text, question_id, questionnaire_id, question_text, contract_commitment_excerpt, questionnaire_answer_excerpt, explanation, severity (HIGH, MEDIUM, LOW), status (UNRESOLVED, DISMISSED, RESOLVED), created_at and updated_at, newest first, plus total, limit and offset. Params: account_id narrows to one account, obligation_id to one obligation (the obligation id shown in contract obligation lists), status to UNRESOLVED, DISMISSED or RESOLVED; limit 1-200 (default 50) and offset. Each finding shows whether what we told a customer in questionnaires matches what their contract commits us to; it carries an excerpt of the commitment, not the full commitment record.

list_conversation_insights

ChatGPT
Paginated, searchable list of chat conversations (the Chat Conversations table). Each row has the question, the answer, source_type (web/slack/etc), answered_at, and the asking user's name/email/profile_image, plus total/page/page_size/total_pages. Answers questions about what people asked in chat; rows carry no citations, agent trace or aggregate counts. Params: page, page_size, q (free-text search over question, answer, user name and email), sort_by/sort_order, source_type, start_date/end_date in microseconds.

list_conversations

ChatGPT
List your recent conversations with Wolfia AI. Returns conversation IDs, titles, and timestamps for resuming previous discussions.

list_integrations

ChatGPT
List all connected integrations and their status. Shows which data sources are configured (Notion, Slack, Google Drive, etc.), how many documents are indexed from each, and when they last synced.

list_missing_information

ChatGPT
The Missing information tab of the Knowledge updates page: knowledge the organization's questionnaires and chats asked for that the knowledge base lacks. gaps holds the knowledge gap report, the topics Wolfia found across many questionnaires, ordered by section, then by term with the most asked term first, then by times_asked, so the most asked gaps across every section are found by sorting every page by times_asked. Each gap has verdict_id, section (AI_GOVERNANCE, PRODUCT or OUTSIDE_PRODUCT), term (the area of the organization's product or company it belongs to), question_type, question_text, missing_knowledge_summary (the knowledge to add), times_asked (distinct questionnaires that asked it), member_question_count, is_waiting_for_approval, business_segment_ids and business_segment_names, representative_question_id, representative_questionnaire_id, created_at and updated_at in microseconds. questionnaire_cards and chat_cards are the missing-information cards from single answers in the last 90 days, listed after the gaps. total counts gaps and cards together and next_offset is the offset of the next page, null on the last page. limit defaults to 50 and can be at most 50, and one page holds at most 20 cards. Closing a gap moves later items to lower offsets, so offsets read before a gap was closed no longer point at the same items. Optional filters: section, term (exact match, as returned on a gap) and business_segment_id; a filtered call returns matching gaps only. Each card has verdict_id, verdict_category, card_type, missing_knowledge_summary, drafted_fact_title and drafted_fact_text (a suggested fact, when one was drafted), conflicting_items (each with item_id, item_kind, item_text, source_name and business_segment_names), old_claim_quote, new_claim_quote, is_waiting_for_approval, account_name for a card about one customer account and created_at in microseconds; a questionnaire card also has questionnaire_id, questionnaire_title, question_id and question_text, and a chat card has conversation_id, message_id and message_snippet. A gap or card closes when a fact is added with its verdict_id as answer_triage_verdict_id, or when it is dismissed. A fact added without a verdict_id leaves every gap and card open. Conflicts between facts and new facts from reviewers' edits are not listed here.

list_new_facts_from_edits

ChatGPT
The What we learned tab of the Knowledge updates page: open cards from the last 90 days where a reviewer's edit to an answer added knowledge the knowledge base does not have. Returns questionnaire_cards and chat_cards, each newest first, questionnaire cards before chat cards across pages. total is how many cards are open and next_offset is the offset of the next page, null on the last page. limit defaults to 20 and can be at most 20. Closing a card moves later cards to lower offsets, so offsets read before a card was closed no longer point at the same cards. Each card has verdict_id, verdict_category, card_type, missing_knowledge_summary, drafted_fact_title and drafted_fact_text (a suggested fact, when one was drafted), conflicting_items (each with item_id, item_kind, item_text, source_name and business_segment_names), old_claim_quote, new_claim_quote, is_waiting_for_approval, account_name for a card about one customer account and created_at in microseconds; a questionnaire card also has questionnaire_id, questionnaire_title, question_id and question_text, and a chat card has conversation_id, message_id and message_snippet. A card closes when the new fact is added with the card's verdict_id as answer_triage_verdict_id, or when it is dismissed. A fact added without a verdict_id leaves every card open.

list_pre_approved_answers

ChatGPT
Paginated list of the organization's pre-approved answers: question and answer pairs the team has signed off for reuse when Wolfia answers questionnaires and chat. Each item has id, question, answer (the company-wide default answer, null when the question only has variants), variants (answers for specific business segments, each with id, answer and tags, the business segment ids it applies to), variant_count, default_answer_tag_ids (the business segments the default answer covers), is_active (inactive answers are never used for answering), is_approved, approved_at and approver, category_ids, citation_count (how many times answering used it), expires_at, internal_note, creator, updater, and created_at and updated_at in microseconds; plus total, limit and offset. Params: q, a case-insensitive text search over the question, the default answer and every variant answer (3 to 500 characters); is_active and is_approved filters; category_id; business_segment_ids, a comma-separated list that keeps answers with a variant in any of those segments or with a company-wide default answer; sort_by (created_at, updated_at, question or citation_count) and sort_order (asc or desc); limit (1 to 1000, default 50) and offset; deleted_only true to list only deleted answers in the trash, in which case q is ignored. Items do not carry original_question, original_answer, a variant's internal_note or original_answer, or monthly match and edit rates.

list_questionnaire_approval_history

ChatGPT
List approval history across your organization's questionnaires, newest first. For each questionnaire returns when it was received, its assignee, total vs approved questions, the approval time-span, per-approver contribution, and an activity window. Timestamps are microseconds since epoch. Use this to measure level-of-effort and forecast how long a new questionnaire will take. Filter with query (title/company substring) or status (e.g. COMPLETED). Page with limit/offset, or window with since_us (inclusive floor) and before_us (exclusive ceiling) on received time. Each record is about 0.7 KB of JSON without question events, so a page of 200 is roughly 130 KB; if your MCP client caps tool output, request smaller pages (limit 40 stays under 30 KB) instead of one large call. Set include_question_events for per-question approval timestamps, which makes records much larger.

list_questionnaire_instruction_versions

ChatGPT
List the version history of your organization's questionnaire answering instructions: each version's number, author, and timestamp, newest first, showing how the answering instructions changed over time. Requires admin or expert role.

list_questionnaires_for_org

ChatGPT
List the latest questionnaires in your Wolfia workspace with their metadata — customer/company name, title, status, question counts, deadline, deal value, assignees, created timestamp, and a page_url link to each questionnaire. Sorted newest first by default; pass sort_by/sort_order as comma-separated lists over company_name, created_at, deadline, deal_value, status and title to order by deal size or deadline instead. Use limit/offset to page. Use this to pull recent questionnaires into other systems or to check what came in recently.

list_reviews_contract

ChatGPT
Paginated list of the legal (contract) reviews shared with the organization. Each row has id, file_name, file_size, status, detected_role/confirmed_role (which side of the contract we are), section_count, obligation_count, is_org_document, is_signed_contract (an executed contract tracked for obligations rather than a draft under review), created_by, created_at and updated_at, plus total/limit/offset. Scoped to org-shared reviews; members' private reviews are never returned here. It finds a specific contract review by name and shows which reviews are still processing; it carries no aggregate counts, findings breakdown or turnaround. Params: sort_by/sort_order (comma-separated over created_at, detected_role, file_name, file_size, section_count, status), a q search over file name/counterparty/status/role, status_filter, detected_role_filter, is_signed_contract, limit (1-100) and offset.

list_sources

ChatGPT
Browse and filter content sources in your knowledge base. Filter by source type, search by name, or filter by business segments. Returns each source's document ID, the same ID that identifies the document when reading its full content or download link. Parameters: - query (string, optional): substring to match against document name or summary. - source_types (array of strings, optional): filter by source type. Valid values: BOX_INTEGRATION, CLARI_COPILOT_INTEGRATION, CONFLUENCE_INTEGRATION, CORE_KNOWLEDGE, CORRECTIONS, DRATA_INTEGRATION, EMAIL_INTEGRATION, GITHUB_INTEGRATION, GITLAB_INTEGRATION, GLEAN_INTEGRATION, GONG_INTEGRATION, GOOGLE_DRIVE_INTEGRATION, GURU_INTEGRATION, HUBSPOT_INTEGRATION, KNOWLEDGE_ANSWER, LINEAR_INTEGRATION, MANUAL_FACT, MINTLIFY_INTEGRATION, MOMENTUM_IO_INTEGRATION, NOTION_INTEGRATION, ONEDRIVE_INTEGRATION, ONELEET_INTEGRATION, PRE_APPROVED_ANSWER, QUESTIONNAIRE_KNOWLEDGE, RIPPLING_INTEGRATION, SALESFORCE_INTEGRATION, SHAREPOINT_INTEGRATION, SLAB_INTEGRATION, SLACK_INTEGRATION, TRUST_CENTER, VANTA_INTEGRATION, WEBSITE_INTEGRATION. - business_segments (array of strings, optional): organization-defined business segment names to prioritize; matching facts rank first and other-segment facts are returned with markers, not filtered out. - limit (integer, optional, default 20): maximum number of sources to return.

list_trust_access_requests

ChatGPT
List trust center access requests with requester name/email, company, request time, status, and decision time. Paginated; filter by status, reason, account, or search.

reactivate_pre_approved_answer

ChatGPT
Let Wolfia use a deactivated pre-approved answer again when it answers questionnaires and chat. pre_approved_id is the id of a pre-approved answer (the pre-approved answer list with is_active false lists inactive answers); there is no body. Sets is_active to true; an answer that is already active is returned unchanged. Deleted answers in the trash cannot be reactivated with this tool. Returns the answer with the same fields as an item of the paginated pre-approved answer list. Errors: 404 when the id does not exist or the answer was deleted. Requires the admin role.

regenerate_questionnaire_answers

ChatGPT
Start AI regeneration of the answers to questions. Takes question_ids and optional instructions to guide the new answers. Regeneration runs in the background; a question is done when its status is no longer ANSWERING_AI. Returns regenerated_count, message, and next_step.

restore_pre_approved_answer

ChatGPT
Bring a deleted pre-approved answer back from the trash, undoing its delete. pre_approved_id is the id of a deleted answer; the pre-approved answer list with deleted_only true lists them. There is no body. The answer comes back with the question, default answer, variants, categories, active status and approval it had when it was deleted. Returns the restored answer with the same fields as an item of the paginated pre-approved answer list. Errors: 404 when the id does not exist or the answer is not deleted; 409 when an answer that is not deleted already has the same question, compared ignoring case and extra spaces (that answer has to be renamed or deleted first). Requires the admin role.

revert_questionnaire_instruction_to_version

ChatGPT
Revert your organization's questionnaire answering instructions to a previous version by version number. This makes that version's content the current instructions and records a new version attributed to you. version_number is a number from the instruction version history. Requires admin or expert role.

save_knowledge_gap_conflict_fix

ChatGPT
Settle a knowledge gap whose card_type is CONFLICT_WITH_KNOWLEDGE_BASE, where facts in the knowledge base disagree with each other or with a reviewer's answer. verdict_id is the verdict_id of an open card on the Conflicts tab of the Knowledge updates page, and the body carries conflicting_fact_ids: one or more item_id values from that gap's conflicting_items whose item_kind is FACT, naming the facts Wolfia should stop using. Send title and fact_text together to save the correct statement as a new manual fact; a gap with a drafted_fact_text needs them, so a statement replaces the hidden facts. On a gap with no draft, send neither to only stop using the chosen facts. The chosen facts are hidden from answering, not deleted from their documents, and the gap is closed. When the organization requires approval for new knowledge, the fix waits for approval and the chosen facts stay in use until it is approved. Returns manual_fact_id (the saved fact, or null when no statement was sent) and update_conflicting_document_url (a link to the page where the document holding a conflicting fact can be edited, or null). Errors: 400 when a fact id is not one of this gap's conflicting facts or the gap has a draft and no statement was sent, 409 when the gap is already closed or already waiting for approval, 404 when the verdict_id does not exist or conflict fixes are not turned on for the organization. Requires permission to edit the knowledge base (an expert or admin role).

search_knowledge

ChatGPT
Semantic vector search across your organization's entire knowledge base. Returns facts grouped by source document with similarity scores. Use for finding relevant information about a topic across all indexed content (Notion, Slack, Google Drive, etc.). Parameters: - query (string, required): natural language search text. - source_types (array of strings, optional): filter by source type. Valid values: BOX_INTEGRATION, CLARI_COPILOT_INTEGRATION, CONFLUENCE_INTEGRATION, CORE_KNOWLEDGE, CORRECTIONS, DRATA_INTEGRATION, EMAIL_INTEGRATION, GITHUB_INTEGRATION, GITLAB_INTEGRATION, GLEAN_INTEGRATION, GONG_INTEGRATION, GOOGLE_DRIVE_INTEGRATION, GURU_INTEGRATION, HUBSPOT_INTEGRATION, KNOWLEDGE_ANSWER, LINEAR_INTEGRATION, MANUAL_FACT, MINTLIFY_INTEGRATION, MOMENTUM_IO_INTEGRATION, NOTION_INTEGRATION, ONEDRIVE_INTEGRATION, ONELEET_INTEGRATION, PRE_APPROVED_ANSWER, QUESTIONNAIRE_KNOWLEDGE, RIPPLING_INTEGRATION, SALESFORCE_INTEGRATION, SHAREPOINT_INTEGRATION, SLAB_INTEGRATION, SLACK_INTEGRATION, TRUST_CENTER, VANTA_INTEGRATION, WEBSITE_INTEGRATION. - business_segments (array of strings, optional): organization-defined business segment names to prioritize; matching facts rank first and other-segment facts are returned with markers, not filtered out. - limit (integer, optional, default 20): maximum number of sources to return. Each returned fact carries a product-relevance marker: [PRODUCT-MATCH-EXCLUSIVE: names] the fact covers only the requested segment(s); [PRODUCT-MATCH-SHARED: names | also tagged: other names] the fact also covers other segments; [PARENT-PRODUCT: name -> parent of selected segment(s)] the fact belongs to the parent of a selected segment; read both segments' descriptions for what the parent shares with the child; when they are silent it is a fallback used only when the child has no fact of its own, stated as the parent's and never as the child's own figure; [CHILD-PRODUCT: name -> child of selected segment] the fact belongs to a child of a selected segment; read both segments' descriptions for whether the child's content answers for the parent; when they are silent it is stated for the parent with the child named, never as the parent's own claim; [SIBLING-PRODUCT: name -> sibling of selected segment] the fact belongs to another child of the same parent; it may only name what that sibling offers so the reader knows which edition has it, and never becomes the selected segment's claim or fills a select value; [OTHER-PRODUCT: names] the fact belongs to a different segment — never present it as applying to the requested one; [ORG-WIDE] the fact applies to the whole organization, except for a requested segment whose description says it inherits nothing or that company-wide facts never answer questions about it; for that segment the fact is not an answer. Other-segment facts are NOT filtered out of these results — they are returned and marked so you can judge them.

search_knowledge_text

ChatGPT
Full-text keyword search across all indexed facts. Use when you know the exact terms to search for (e.g., 'SOC 2', 'deployment process', 'rate limiting'). Returns facts grouped by source document with text relevance scores. Parameters: - query (string, required): exact keywords or phrase to search for. - source_types (array of strings, optional): filter by source type. Valid values: BOX_INTEGRATION, CLARI_COPILOT_INTEGRATION, CONFLUENCE_INTEGRATION, CORE_KNOWLEDGE, CORRECTIONS, DRATA_INTEGRATION, EMAIL_INTEGRATION, GITHUB_INTEGRATION, GITLAB_INTEGRATION, GLEAN_INTEGRATION, GONG_INTEGRATION, GOOGLE_DRIVE_INTEGRATION, GURU_INTEGRATION, HUBSPOT_INTEGRATION, KNOWLEDGE_ANSWER, LINEAR_INTEGRATION, MANUAL_FACT, MINTLIFY_INTEGRATION, MOMENTUM_IO_INTEGRATION, NOTION_INTEGRATION, ONEDRIVE_INTEGRATION, ONELEET_INTEGRATION, PRE_APPROVED_ANSWER, QUESTIONNAIRE_KNOWLEDGE, RIPPLING_INTEGRATION, SALESFORCE_INTEGRATION, SHAREPOINT_INTEGRATION, SLAB_INTEGRATION, SLACK_INTEGRATION, TRUST_CENTER, VANTA_INTEGRATION, WEBSITE_INTEGRATION. - business_segments (array of strings, optional): organization-defined business segment names to prioritize; matching facts rank first and other-segment facts are returned with markers, not filtered out. - limit (integer, optional, default 20): maximum number of sources to return. Each returned fact carries a product-relevance marker: [PRODUCT-MATCH-EXCLUSIVE: names] the fact covers only the requested segment(s); [PRODUCT-MATCH-SHARED: names | also tagged: other names] the fact also covers other segments; [PARENT-PRODUCT: name -> parent of selected segment(s)] the fact belongs to the parent of a selected segment; read both segments' descriptions for what the parent shares with the child; when they are silent it is a fallback used only when the child has no fact of its own, stated as the parent's and never as the child's own figure; [CHILD-PRODUCT: name -> child of selected segment] the fact belongs to a child of a selected segment; read both segments' descriptions for whether the child's content answers for the parent; when they are silent it is stated for the parent with the child named, never as the parent's own claim; [SIBLING-PRODUCT: name -> sibling of selected segment] the fact belongs to another child of the same parent; it may only name what that sibling offers so the reader knows which edition has it, and never becomes the selected segment's claim or fills a select value; [OTHER-PRODUCT: names] the fact belongs to a different segment — never present it as applying to the requested one; [ORG-WIDE] the fact applies to the whole organization, except for a requested segment whose description says it inherits nothing or that company-wide facts never answer questions about it; for that segment the fact is not an answer. Other-segment facts are NOT filtered out of these results — they are returned and marked so you can judge them.

search_similar_pre_approved_answers

ChatGPT
Find the pre-approved answers whose question means the same as a given question, ranked by semantic similarity to question, most similar first. Shows whether the library already answers a question, and finds duplicates. Body: question (required, 1 to 2000 characters). Param: limit, 1 to 10 (default 5). Only active answers that are not deleted are returned; approved and unapproved answers are both included. Returns items, each with item (the full pre-approved answer with the same fields as an item of the paginated pre-approved answer list, including id, question and the company-wide default answer; business segment variants are not included and appear in the full detail of the answer) and similarity_score (cosine similarity between the two questions, higher is closer, 1.0 is identical). An empty items list means no active answer exists or the question could not be embedded. This is a meaning search over questions, not a keyword search over question and answer text. Requires an expert or admin role.

search_slack_messages

ChatGPT
Search raw Slack messages by text content. Returns matching messages with sender names, channel names, and direct thread links. Optionally filter to a specific channel by name.

send_message

ChatGPT
Send a message to Wolfia AI and get a response with citations. Answers questions about the organization's knowledge base, documents, policies, and internal information. Both modes answer from the approved knowledge base and include sources. mode='fast', the default, answers in seconds, including detailed, multi-part, and customer-facing questions. mode='thinking' is a deeper, slower, exhaustive review: it verifies each claim against its sources and takes several minutes. Progress notifications stream while the answer is generated, and the first one carries the conversation_id. If the call times out on the client side, the answer is still generated and stored under that conversation_id, where the conversation history returns it.

submit_questionnaire

ChatGPT
Submit a security questionnaire to Wolfia for extraction and automated answering. Takes exactly one of: text_content (questions pasted from Slack, email, or a portal); file_content (base64-encoded file bytes, with file_name set) to submit a file in a single call; or file_upload_id (the id returned when a questionnaire file upload URL is created, once the file has been PUT to that URL). A large base64 file_content argument can stall the tool call, so file_content suits small files. Supported file types: xlsx, xlsm, xls, docx, pdf, csv, png, jpg, and zip. A zip creates one questionnaire per questionnaire file inside it, listed under questionnaires in the response. Optional fields mirror the upload form: company_name (the account this questionnaire is for), instructions (extra guidance for answering), deadline (ISO date like 2026-07-01), deal_value (a number like 100000), business_segments (a list of segment names from your workspace), crm_opportunity_id (a Wolfia CRM opportunity UUID), crm_account_id (the CRM account's external record id, e.g. a Salesforce Account Id — resolved automatically from the opportunity when only that is given), and assignee_emails (a list of team-member email addresses to assign this questionnaire to — every email must match a user in your workspace or the submission is rejected). Returns the questionnaire ID and a link to track answering progress.

unapprove_pre_approved_answer

ChatGPT
Remove the approval from one pre-approved answer. pre_approved_id is the id of a pre-approved answer (the pre-approved answer list with is_approved true lists approved answers); there is no body. Sets is_approved to false and clears approved_at and the approver; an answer that is not approved is returned unchanged. When the organization requires approval before pre-approved answers are used, the answer stops being used in answering until it is approved again. Returns the answer with the same fields as an item of the paginated pre-approved answer list. Errors: 404 when the id does not exist or the answer was deleted. Requires an expert or admin role.

update_contract_obligation_renegotiate_flag

ChatGPT
Flag or unflag one signed-contract obligation to be renegotiated to our standard terms at the customer's renewal, and record the change in the obligation's activity history. obligation_id is the obligation id shown in contract obligation lists. Body: renegotiate_to_standard (required, true to flag, false to remove the flag). Sending the value the obligation already has still adds an activity entry. The flag applies to bespoke terms that deviate from our standard paper, as described by an obligation's standard_term_deviation_summary. Returns the updated obligation, with renegotiate_to_standard set to the new value and the same fields as the full obligation record except counterparty_name. Errors: 404 when the obligation_id does not exist; 422 when renegotiate_to_standard is missing. Requires the admin role.

update_contract_obligation_status

ChatGPT
Move one signed-contract obligation to its next tracking status and record who did it in the obligation's activity history. obligation_id is the obligation id shown in contract obligation lists (for a merged row naming several customers, each customer's own obligation_id is in member_obligations). Body: status (required, one of ACKNOWLEDGED, IN_PROGRESS, COMPLETED, VERIFIED, WAIVED) and note (optional text saved on the activity entry). Allowed moves: IDENTIFIED to ACKNOWLEDGED or WAIVED; ACKNOWLEDGED to IN_PROGRESS or WAIVED; IN_PROGRESS to COMPLETED or WAIVED; COMPLETED to VERIFIED or WAIVED. VERIFIED and WAIVED are final, and a status cannot be set to the one it already has, so the allowed moves depend on the obligation's current status. Requires an expert or admin role, and moving to VERIFIED requires the admin role. Returns the updated obligation with the same fields as the full obligation record except counterparty_name. Errors: 400 for a move not in the allowed list, such as IDENTIFIED to COMPLETED; 403 when a caller without the admin role asks for VERIFIED; 404 when the obligation_id does not exist; 422 for a status outside the five values.

update_knowledge

ChatGPT
Update an existing knowledge-base fact by its id. Change the title, fact_text, context, business-segment tag_ids, or shareability. Requires an expert or admin role.

update_pre_approved_answer

ChatGPT
Edit one pre-approved answer, by pre_approved_id (the id of a pre-approved answer). Body, send at least one of: question (new question text, not empty); answer (new company-wide default answer, not empty); category_ids (replaces the whole list of category ids; an empty list removes every category); internal_note (a note for reviewers that answering never uses; an empty string clears it). Optional answer_triage_verdict_id: the verdict_id of an open knowledge gap on the Missing information tab of the Knowledge updates page that this edit fixes; the gap is closed after the save. Omitted fields are left unchanged, and question or answer sent as null are also left unchanged, so this tool cannot clear the default answer. The first edit keeps the earlier text in original_question and original_answer. Approval and active status do not change. Variant answers are not edited here. Returns the updated answer with the same fields as an item of the paginated pre-approved answer list. Errors: 400 when none of question, answer, category_ids or internal_note is sent; 404 when the id does not exist, the answer was deleted, or the new question matches another answer's question; 404 when answer_triage_verdict_id does not exist; 409 when that knowledge gap is already closed or waiting for approval; 422 when question or answer is an empty string. Requires an expert or admin role.

update_trust_portal_settings

ChatGPT
Update your organization's Trust Center (trust portal) settings: access-request rules, NDA/watermark config, domain blocking, and customer-upload options. Pass only the fields to change; omitted fields keep their current values. Requires admin role.

update_user_correction

ChatGPT
Update an existing correction by its id, changing its title or corrected answer text. Requires an expert or admin role.

update_variant

ChatGPT
Edit one business segment variant of a pre-approved answer. pre_approved_id is the id of the pre-approved answer and variant_id is the id of an entry in its variants. Body, send at least one of: answer (new text, not empty); tags (replaces the variant's business segment ids; one or more, each ACTIVE and not in another variant of this answer); internal_note (a note for reviewers that answering never uses; an empty string clears it). Omitted fields, and fields sent as null, are left unchanged. The first answer edit keeps the earlier text in original_answer. The company-wide default answer is not changed here. Returns the variant: id, pre_approved_answer_id, answer, internal_note, original_answer, tags, created_by, created_at and updated_at in microseconds, updated_by. Errors: 400 when no field is sent; 404 when the variant does not exist, does not belong to pre_approved_id, its pre-approved answer was deleted, a business segment is not ACTIVE, or a segment is already in another variant of this answer; 422 when answer or tags is sent empty. Requires an expert or admin role.

upsert_editorial_guide

ChatGPT
Create or replace your organization's editorial guide (the writing-style and tone rules the answering agent follows). Pass the full guide body; this overwrites the whole current guide, so text left out of the body is removed, and records a new version. Requires admin or expert role.

upsert_slack_auto_respond_instructions

ChatGPT
Create or replace your organization's Slack auto-respond instructions (org-level guidance for auto-answering in Slack, the insufficient-knowledge fallback message, and the respond-once-per-thread setting). This overwrites all of the current instructions, so anything left out is removed, and records a new version. Requires admin or expert role.

Capabilities

Writes

Example Prompts

Click any prompt to copy it.

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/wolfia"><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/wolfia)