Risk matrix

MCP is the standard that lets an AI assistant — a general-purpose chat assistant, for example — READ your company's context here inside to answer your questions, without you having to copy and paste any report. On this screen you create the key the assistant will use, choose what it can read and see when it last read. The question the screen answers is: "what can an outside AI see of my company, and who authorized it?".

No MCP scope writes, changes or deletes any of your data. The assistant reads what you've released and answers; every change keeps happening through the screen, with your login and your audit trail. It's a design decision, not a setting: there's no button to "let the AI write" on this screen.

Whoever has the key has its scopes, without needing your login. Don't paste the key into a group channel, issue, shared spreadsheet or repository. If you suspect exposure, revoke it immediately: revocation takes effect at once and the "last used" column shows whether the key was used after you became suspicious — that's the information that closes (or opens) an incident.

Copy the connector address from this screen and paste it in the store: in Claude, Settings › Connectors › Add custom connector; in ChatGPT, Settings › Apps & Connectors › Create, in developer mode, with OAuth authentication; in Copilot Studio, Tools › Add a tool › Model Context Protocol. The person signs in with their own Tenka login. The first time, the app shows up as pending in Connected apps, and whoever administers the account approves it. Store connections only work once the platform opens the stores for the account; until then, the API key on this screen still works.

Three steps. (1) Here, create the key and tick the scopes — start with only what you need; you can widen it later. (2) The key appears ONCE on screen, at the moment it's created: copy it then, because the platform only keeps its cryptographic digest and can no longer tell what it was. (3) Paste the key into your assistant's MCP configuration. If you lose the key, there's no recovery — revoke it and create another, which takes ten seconds.

Each scope is a slice of context, and the name says which slice: `empresa.ler` (identity, industry, structure), `contrato.ler` (portfolio and contracted amounts), `demanda.ler` (requests, status, time entries) and `financeiro.resumo` (totals and indicators — summary, not entry by entry). Good practice is the usual: release the minimum that answers your question. An extra scope doesn't improve the answer and widens what leaks if the key is exposed.