ExactCov on Claude and ChatGPT
Your analysts already work in Claude or ChatGPT. Now the covenant register is one question away: which names are near a trigger, what a basket allows, which certificates are late. Every answer is cited to the clause and page, and nothing you ask is retained.
在 Claude 或 ChatGPT Enterprise 中直接询问 ExactCov
ExactCov 以 MCP 服务器和 Skills 的形式提供。您的团队可在已有的 Claude 和 ChatGPT Enterprise 订阅中提出契约问题，每个答案都来自登记册并附带引用。权限同步，零数据保留。
The register, in plain words, with the citation
Which names are near a trigger?
Headroom under a band you name, filtered by test, sponsor, sector or reporting status. Every row with its clause and page.
What can they still raise?
Day-one capacity under the debt and lien baskets for one name, with each component and the definition it rests on.
What has not arrived?
Deliverables overdue across the book, how many times each was chased, and a draft of the next chase for you to approve.
What changed this week?
Tests that moved, documents that landed, waivers in progress, in the order the morning brief would put them.
Show me the clause
The exact provision behind any figure, quoted, with the page. Definition chains walked hop by hop on request.
Draft the memo
A committee note or a chase email written from register facts only. The assistant drafts, you edit, a person sends.
An MCP server and Skills, not a copy of your data
ExactCov exposes the register through the Model Context Protocol, the open standard both Claude and ChatGPT use to connect to tools, and ships Skills that teach the assistant how to ask well. The assistant calls the register at question time. Nothing is exported, indexed or copied into the assistant.
- Permissions mirror ExactCov. A user sees exactly the facilities they see in the register, nothing more.
- Every answer carries the citation the register holds. If the register cannot cite it, the assistant does not say it.
- Actions are proposals. Chase emails, escalations and memos are drafted for a person to send.
- Zero data retention applies end to end. Questions and answers are not used to train any model.