Transcript
A key is only worth what the agent does with it. Here is the same workspace, one question, and the trail it leaves behind.
Where we left off
In the previous video we created an access key for one agent. Now we use it.
Handing the key to the agent
Our example agent is Claude, but nothing here is specific to it — any MCP client works the same way. The agent needs two things from DocReveal: the access key and the workspace address. How you store them inside your agent is covered by that agent's own documentation, so we go straight to the question: summarise the invoice named invoice one.
The agent connects
The agent opens a session against DocReveal and loads the actions your key allows. Anything outside that key is not part of the list it receives.
Finding the document
It lists the documents in the workspace, finds the invoice and reads its profile. Give this part a moment — the agent is making several calls, and each one runs against your data, not against a cached copy.
The answer
The answer comes back from the document itself: invoice number, issue date, both parties, the line items, and the totals — net, VAT and gross. No template was defined for this invoice in advance.
Every call is logged
Back in DocReveal, the Activity tab shows what actually happened. Nothing is waiting for your approval, because this agent only used read actions. Below that, every call is recorded with its time, the action, the capability it belongs to, the key that made it, and how long it took. The audit trail is per key — so when several agents share a workspace, you still know which one did what.
Back to the source
An answer you cannot verify is not worth much. The document the agent used is in File Manager, where you can open it.
Checking the answer
Same invoice number, same parties, same totals. The agent read your document, and you can prove it.
Connect an agent, scope it, and keep the receipts. That is agentic AI you can put in front of an auditor.