Can I trust this request?
A rehearsal of the Leagle.id MCP, in a terminal. Pick your kind of company and see why “who authorised this?” beats “is it a bot?”.
Why “is it a bot?” stops helping
Agents already book, buy and manage accounts on people’s behalf. Some are scripts you should keep out. Many are someone’s assistant, and you would like to serve them. A bot check cannot tell the two apart, and it should not have to.
What you needed all along was never “is this a person?” It was “did a person authorise this, and for what?” Some companies only need the first half, that one real person is there. The same signed proof answers both.
What you would want to know instead
Who authorised this agent, for which action, until when, and can you prove it later without having collected their documents? That is a signed chain: a person grants a right, each link can only narrow it, and expiry can only shrink.
You see a pseudonym, keep the certificate as your record, and can ask the person to account for a delegate if something goes wrong. Nobody has to hand over a password, and you do not have to hold their identity to be sure.
What this page is
A rehearsal. The terminal below is a preset conversation, not a live model. It follows the order the MCP follows with a real project: explore, then license, then scaffold or wire in. Pick your kind of company and answer a few questions; the replies are what an agent would tell you from the facts the MCP carries.
What it is not
It is not a sandbox. Nothing is generated or executed, and where an agent would write code in your repository, it says so. If your company is not in the list, the last option shows how to ask the MCP about your own code instead.
$ claude mcp add leagle-id -- npx -y @leagle/cli mcp
$ claude
Who are your customers?