Ocolta

View saved version →

---
name: ocolta-review-document-signals
description: "Explain categorical signals from an existing purchased Ocolta document review, compare accessible summaries, or recover a recent result without exposing document text."
---

Use the Ocolta MCP at https://app.ocolta.com/mcp through normal OAuth in the purchasing browser. If unavailable, give the integration handoff at https://app.ocolta.com/workspace/integrations. This workflow reads existing results only.



# Review document signals

Call `get_connected_purchase` to identify the linked purchase and `list_recent_reviews` to find recoverable results. Use only IDs returned for that owner. Read the selected result with `get_review_summary`, preserving status, confidence, priority and indicator counts.

Translate the returned categories into a concise review summary and questions the user can investigate. Do not infer facts absent from the categorical response, reconstruct raw document text, expose sensitive identifiers, verify a bank account, or decide a person's eligibility. Confidence is not a guarantee that a document is authentic.

For a new user or an empty result list, explain that this connection reads existing purchased reviews; it cannot upload a document or buy a review. Use the returned product link when provided, otherwise point to `https://app.ocolta.com` to begin the normal review flow. Do not claim the handoff completes a purchase or analysis. Linking must use the purchasing browser that owns the review.

For a return visit, refresh the accessible review list before interpreting an old ID. The product's recovery window is currently 24 hours; an expired or missing result is not a clean assessment. Compare only still-accessible categorical results, keeping their separate identifiers and timestamps. Do not promise durable history or monitoring.

View saved version →

---
name: ocolta-verification-follow-up
description: "Prepare neutral human follow-up questions from an Ocolta categorical result or user-described concern, without deciding authenticity or eligibility."
---

# Plan source verification follow-up

Use an actual Ocolta result or the user’s description of the concern. A connected purchase at https://app.ocolta.com/mcp can be identified with `get_connected_purchase`; use `list_recent_reviews` and `get_review_summary` for a recoverable owned result. Recovery is currently 24 hours after completion. If unavailable, say the plan is based on supplied information and give https://app.ocolta.com/workspace/integrations as the connection handoff.

Preserve categorical status, priority, confidence, indicator counts and limitations. These fields do not expose the underlying document text or locate individual indicators; the user reviews the full report and source in the purchasing browser. Do not invent exact findings or reconstruct personal fields. An inconclusive or refused result supplies no authenticity determination, and a low-priority result does not prove a document is valid.

Ask only for material context needed for a useful checklist, such as document type, decision being reviewed and the organization’s existing verification procedure. The checklist does not need account numbers, identity numbers, credentials or banking/payroll login details.

Propose checking the source report against the actual file, considering ordinary explanations for artifacts, requesting an untouched original or appropriate adjacent period when needed, and verifying material facts through independently sourced institution contacts under the user’s applicable authorization procedure. Do not use an unverified contact printed on a questioned document as independent proof. Preserve uncertainty if the required source is unavailable.

Draft neutral questions or a request for a clearer source only when asked. Keep observations and proposed checks separate from confirmed facts. The user follows their established review policy and records what was actually confirmed. This workflow does not contact anyone, verify account ownership, accuse a person, establish fraud/AI generation, or recommend approval, denial or reporting. A categorical result alone does not support such a decision.