MCP access tokens

Credentials that let an outside AI assistant reach this operator's data. Each one authenticates a model — not a person — confined to one operator and the tools named on it.

← Agent proposals

Tokens

Your session has expired. Sign in again.

This is the list being unreachable, not there being no tokens. Do not conclude that nothing holds a credential.

USABLE NOW0Active and not past their expiry

LAPSED0Still ACTIVE but past expiry. They are refused, and worth tidying.

REVOKED0Kept, so the tool-call history still names what made each call

NEVER USED0Issued and never called. Not evidence of anything, but worth a look.

Issued tokens

The hint is the last few characters of the secret, which is all the platform kept. There is no column here that could show the token itself.

No MCP token has been issued for this operator. No outside assistant can reach this data.

Issue a token

The secret appears once, here, and is never served again.

Issuing a token requires mcp.token.write, which your role does not carry in this scope. The permission is human-only by design: a model must never be able to mint a credential for itself. The form is not shown because it would only reach a refusal — but the API is what enforces this, not this screen.