Authentication
An MCP server has two independent layers of authentication, and keeping them separate makes everything else clearer:
- Who is allowed to connect to the MCP server? This is Scalar-side access control—public, your team, or an access group.
- How does the server call your upstream API? This is upstream auth, configured per installation: a credential you store, or a credential the caller passes through.
They are independent. Someone can be allowed to connect (layer 1) while the server calls your API with a key you stored—or with a key they supply themselves (layer 2).
Who can connect
Every installation is private by default. There are three ways a caller can be granted access:
| Access | Who gets in | How they authenticate |
|---|---|---|
| Public | Anyone with the URL | No Scalar auth |
| Team | Members of the installation's owning team | Personal Access Token or OAuth |
| Access group | Any email/domain on the allowlist—no Scalar account required | OAuth login (email or SSO) |
How the server authenticates to your API
Separately from who connects, each installation decides how it authenticates against your upstream API:
- Global — You store one credential on the installation and the server uses it for every call. Agents never see it.
- Passthrough — The caller supplies the credential, and Scalar forwards it upstream per request without storing it.
Pick a recipe
Most setups are one of these three combinations:
- Public MCP with passthrough auth — anyone can connect, and each user authenticates with their own API credentials. The simplest way to share one MCP with the world.
- Private access for customers — gate the server to specific emails or domains with access groups and OAuth login, and brand the sign-in page.
- One shared key for everyone — store a single credential on the installation so callers never handle a key.
Related
- MCP Servers — create a server and connect a client
- Getting Started — from OpenAPI to agent-ready
