Credentials for MCP, CLI and API

Published

Reviewed

πŸ”‘

Two keys, three doors.

Everything outside the dashboard authenticates with one of two credentials: a browser sign-in or a personal API token. Which one a tool accepts, and what it may do there, differs per tool.

About credentials

A browser sign-in starts in the MCP client or with frbit auth login. A person approves it in the dashboard, and the tool stores and renews it. An API token is created in the dashboard and pasted where needed.

A browser sign-in works only for the tool that started it. An API token works everywhere.

Where each credential works

CredentialMCP serverCLI and public API
MCP sign-inYesNo
CLI sign-inNoYes
API tokenYesYes

Agent skills hold no credential of their own. They use the MCP server or the CLI.

What each credential may do

CredentialReadWriteDelete
MCP sign-inYesYesNot offered
CLI sign-inYesYesWhen allowed
API token, readYesNoNo
API token, read and writeYesYesNo
API token, read, write and deleteYesYesYes

Write means create and change. Over MCP, an agent creates apps and environments, changes environment variables, deploys and restarts. No MCP tool deletes, so the delete permission changes nothing there.

API token permissions apply on the MCP server too: a read-only token calls the read tools only.

The CLI sign-in deletes only when Allow deletes was ticked on approval. Restoring a backup and switching a component off count as deletes, see token permissions.

Lifetime and revocation

Browser sign-inAPI token
CreatedMCP client, frbit auth loginDashboard
Expires1 year after approval, or after 30 days unusedNever
RevokeAI agents in the account menuAPI tokens

Which one to use

  • A person at their own computer: the browser sign-in.
  • A CI job, a script or a server: an API token with the smallest permission that covers the job.
  • An MCP client without OAuth support: an API token, read-only unless the agent should change things.

Written by humans and AI in collaboration.

AI use & editorial processEdit on GitHub β†—