Remote MCP Authentication: A Builder's Checklist

By AHMAD ·

Check token audience, scopes, transport security and permission boundaries for a remote MCP server.

Remote MCP Authentication: A Builder's Checklist

A remote MCP server can connect an AI host to private data or write operations. Authentication proves who is calling; authorization decides what that caller may do. The MCP authorization security considerations (https://github.com/modelcontextprotocol/modelcontextprotocol/blob/main/docs/specification/2026-07-28/basic/authorization/security-considerations.mdx) require servers to validate that a token was issued for the server itself, and the official authorization tutorial (https://github.com/modelcontextprotocol/modelcontextprotocol/blob/main/docs/docs/2026-07-28/tutorials/security/authorization.mdx) warns against passing a token through to a different API.

Review the boundary

For each tool, record the user identity it acts for, the service it calls, the scope it needs, and the data it can return. A token for one service should not become a universal credential for every downstream service. Use separate credentials for downstream APIs and enforce their scopes in code.

Check the transport

Use HTTPS for the remote authorization path. Validate redirect targets and token audience as the specification requires. If you operate a proxy, follow the specification's confused-deputy guidance rather than assuming that a valid login alone protects third-party data.

Test denials

Try a token issued for another audience, an expired token, a user without the required scope, and a write attempted with a read-only role. Each should fail before the underlying action runs. Log the decision and target, but keep tokens out of logs. Test with the actual client and server versions you deploy.

A permission check in a prompt is not an authorization layer. The server and backend API must reject calls that exceed the caller's rights. For broader agent controls, see our security checklist (https://ubits.tech/learning/p/ai-agent-security-checklist-prompt-injection-tool-access).