varsa_verify_claims
ChatGPT[batch version of varsa_verify_response] Verify a pre-extracted list of claims (one per array entry). Runs in the background like varsa_verify_response: SPEAK the returned user_message, then poll varsa_verify_response_status with the job_id until verdicts are ready. Scope: veriref:verify.
varsa_verify_response
ChatGPTVerify factual claims in a passage against authoritative sources. Use when the user asks to verify, fact-check, or find citations for written claims. If the user submits distinct wording of a previously checked claim (including a dropped or restored qualifier), call this tool again rather than reusing a prior receipt from memory. Checks each sentence against peer-reviewed literature (PubMed, Europe PMC), regulatory records (FDA labels, ClinicalTrials.gov, EMA, DailyMed, EMDB cryo-EM depositions), financial filings (SEC EDGAR, FRED, ECB, Eurostat, OECD, Companies House), FASB Accounting Standards Updates (curated ASU registry for GAAP/NFP prompts), insurance regulators (NAIC, CMS), court opinions (CourtListener), legislation (eCFR, Federal Register, EUR-Lex, legislation.gov.uk), patents (USPTO), supplementary web search, and the user's approved reference repository. URLs the assistant cited in the passage are surfaced as assistant-cited links (marked †) even when primary retrieval does not anchor. Returns a per-claim verdict with citations and source links, and saves references to the user's VarsaAI library. Auto-detects the relevant vertical (life_sciences, finance, insurance, or legal) from content — pass a specific verticals subset to scope (e.g. ["finance"] for FASB/GAAP, ["life_sciences"] for EMDB/structural biology; avoid passing unrelated verticals). unsupported means not anchored in connected sources, not that the claim is false. Requires Advanced or Pro Referencing subscription. Scope: veriref:verify.
varsa_verify_response_status
ChatGPTPoll a varsa_verify_response job that returned status: pending. Pass the job_id from that response. Returns the same markdown + verdict payload as varsa_verify_response once verification finishes. Call every 5-15 seconds until status becomes succeeded (or failed). If the response says status: stalled, STOP polling, tell the user the verification is stuck, and ask them to reconnect the VarsaAI connector (Claude: Settings → Connectors → VarsaAI → Reconnect; ChatGPT: Settings → Apps → VarsaAI → Reconnect) and retry in a fresh chat. Scope: veriref:verify.
veriref_approved_claims_list
ChatGPT[read-only] List MLR-approved claims in the caller's VeriRef memory. These are sentences a human reviewer signed off on via the in-app review panel; their embeddings power the semantic recall step in veriref_verify_run — when a new claim is ≥0.90 cosine-similar to an entry here, it's returned as verdict: human_approved without re-hitting external sources. Includes hit_count and last_hit_at. Scope: veriref:approved:read.
veriref_export_run
ChatGPT[read-only export] Export an MLR-ready package. Pass projectId for a VeriCore video project (accuracy_audit refs) OR verificationId for a standalone varsa_verify_response / veriref_verifications row. JSON/CSV/Veeva include each claim's verdict; a contradicted claim always carries its conflicting passage. Formats: json/mlr, veeva, csv. Scope: veriref:export or veriref:verify.
veriref_library_list
ChatGPT[read-only] List saved references in the caller's VeriRef library. scope=user returns the personal store, scope=org returns the shared team store, scope=all (default) merges both. Supports full-text search over titles/notes and tag overlap filter. Every returned reference is the exact ProjectReference shape (with numbered n, url, pmid/nct_id/doi where applicable) so results can be re-attached to a new project. Scope: veriref:library:read.
veriref_library_search
ChatGPT[read-only] Search the caller's VeriRef reference library by keyword and/or vertical. Returns normalized quote, deep_link (alias for locator_url), and trust_tier on every hit. Alias for veriref_library_list with a search-first interface — use when finding specific references rather than browsing. Scope: veriref:library:read.
veriref_receipt_get
ChatGPT[read-only] Fetch a signed VeriRef receipt by id. Returns the canonical claim + references + Ed25519 signature payload, plus a public verifier URL any third party can hit (the response carries the exact link, and the signing public key is published alongside it). Scoped to the caller's own receipts (user or org). Scope: veriref:read or veriref:verify.
veriref_report_get
ChatGPT[read-only · chat-ready] Convenience wrapper that returns ONLY Markdown — no JSON noise — for a stored VeriRef verification. Ideal for ChatGPT/Claude when the user says 'show me my last VeriRef report' or 'render <filename> as a report'. Style options: inline (sentence blocks + References footer, default) or bibliography_only (just the deduplicated References list). Scope: vericore:read or veriref:verify.
veriref_session_diagnose
ChatGPT[read-only · no scope required · DIAGNOSTICS] Returns diagnostic info for this Referencing MCP session: granted scopes, module binding, subscription status, and tool visibility. Works without an active plan — call this first when a trial or unpaid account hits billing errors. Use when the user reports 'not all permissions were granted', connection issues, or missing tools.
veriref_share_create
ChatGPTCreates a PUBLIC link to a verification report: anyone with the URL can view it until it expires (default 30 days, max 30) or is revoked. Only call when the user explicitly asks to share a report. Tell the user the link is public, its expiry date, and that it can be revoked with veriref_share_revoke. Returns the share URL, link id and expiry.
veriref_share_public_get
ChatGPT[read-only][public] Fetch a publicly-shared VeriRef verification by share token. The token IS the auth — no MCP scope is checked here, but the token must be active (not revoked, not past expires_at). Throws not_found | revoked | expired. Closed links never include the verification.
veriref_signoff_request
ChatGPTCreates a pending review (sign-off) request in VarsaAI for selected sentences. If assigneeEmail is provided, that reviewer will be able to see the sentences and note. Only call when the user explicitly asks to request a sign-off or review. Returns the created request ids, sentences, status and assignee.
veriref_verification_get
ChatGPT[read-only] Fetch a single VeriRef verification by id OR by filename substring (most recent match wins). Returns the full row including sentence-by-sentence results[] (each with refs[] carrying PMID/source badges). Set format:'markdown' (or 'both') to also get a chat-ready markdown string (sentence blocks + global References footer) — ideal when the user says 'show me my last VeriRef report'. Scope: vericore:read or veriref:verify.
veriref_verifications_list
ChatGPT[read-only] List the token user's VeriRef verifications (personal + every org), newest first. Use this BEFORE veriref_verification_get when the user says 'my latest VeriRef check' or names a filename. Scope required: vericore:read or veriref:verify.
veriref_verify_run
ChatGPT[single-source · PubMed only] Sentence-level PubMed lookup on a passage. Use when the user asks to verify medical or scientific claims against PubMed specifically. Returns up to 3 citations per sentence with support status, confidence score, and clickable links. Persists a veriref_verifications row so the result can be re-rendered later with veriref_verification_get or veriref_report_get. For broader source coverage (FDA labels, ClinicalTrials.gov, Europe PMC, repository), use veriref_extract_run on the same text. The response markdown field contains the full report with a ## References footer. Counts against MCP quota. Scope: veriref:verify (or veriref:*).