MCP authorization: OAuth for tool servers
How an MCP server reached over HTTP asks for tokens, why tokens must be issued for that server, and what this means for an account-data server.
The MCP specification defines an authorisation framework for servers reached over HTTP. It is optional for MCP as a whole, but implementations using an HTTP transport should follow it, while servers on a local standard-input transport should take credentials from the environment instead.
The roles come from OAuth. A protected MCP server acts as an OAuth 2.1 resource server, the MCP client acts as an OAuth 2.1 client on behalf of the resource owner, and an authorization server deals with the user and issues tokens. The specification builds on the OAuth 2.1 draft and several RFCs, including Protected Resource Metadata, which every MCP server must implement so clients can discover where to get tokens.
Two rules matter most for finance:
- Audience. MCP servers must check that an access token was issued specifically for them, must accept only tokens valid for their own resources, and must not accept or pass through any other token. A token for the account-data server is not a token for the payments server.
- Least privilege. Servers should say which scopes an operation needs, and clients should request only those, asking for more through a step-up flow when needed.
In a finance flow, a bank's MCP server can rely on its existing authorization server, issue narrowly scoped tokens per tool, and refuse anything issued for another audience.
Limits. A scope says what kind of call is allowed, not the amount, payee or deadline of a specific payment. Those details need a mandate or rich authorization data on top.
Sources: Model Context Protocol specification: Authorization (version 2026-07-28), RFC 6749: The OAuth 2.0 Authorization Framework, RFC 9396: OAuth 2.0 Rich Authorization Requests