Auth0MCP Server

Auth0's official MCP server connects an agent to an Auth0 tenant, so identity work — creating applications, deploying Actions, managing users, and reading logs — happens where you are already working instead of in the dashboard.

What it is for

The operations it targets are the ones that interrupt development: standing up a new application and retrieving its domain and client ID, writing and deploying an Action, checking whether a login attempt from a specific IP succeeded, or looking up why a user is failing authentication. Each is a short dashboard round trip on its own; together they add up, and they are exactly the kind of task an agent can complete from a sentence.

Scoped access by default

The --tools flag decides which tools are exposed, and it is the important knob. npx @auth0/auth0-mcp-server init --read-only gives an agent inspection without mutation; a glob such as --tools 'auth0_list_*,auth0_get_*' selects explicitly. For a system that holds your identity provider's management credentials, starting read-only and widening deliberately is the sane default.

Client setup

An init subcommand configures the server for a specific client — Claude Desktop, Claude Code, Windsurf, and Cursor all have first-class paths. For Claude Code you choose a scope: user, written to ~/.claude.json and available across projects, or project, written to .mcp.json so it can be committed and shared with your team.

Status

The server is published under the MIT licence and is explicitly beta software, provided as-is: features and APIs may change without notice, support during the beta is limited, and Auth0 does not recommend it for production or critical workloads. Treat it as a development-time accelerator against a non-production tenant rather than something to wire into a release pipeline. Requires Node.js 18 or later.