A high-severity vulnerability (CVE-2026-115563) in the Model Context Protocol (MCP) Python SDK allowed a malicious or compromised MCP server to redirect a client's OAuth credentials to an attacker-controlled token endpoint, according to a GitHub security advisory published November 191. The flaw affected mcp versions 1.9.1 through 1.29.x and 2.0.0a1 through 2.1.x; patches are available in versions 1.30.0 and 2.2.0.
The MCP Python SDK's OAuth client support failed to validate the authorization server metadata issuer on every discovery path and did not bind stored or pre-provisioned client credentials to the authorization server they belonged to. A malicious or compromised MCP server could exploit either gap to direct credentials to a token endpoint of its choosing — by naming its own authorization server in its protected resource metadata, or by publishing none and serving metadata that presented the user's real authorization server as the issuer. The server would then receive the client_secret, authorization code, and PKCE code_verifier meant for the legitimate server. The advisory notes that with PrivateKeyJWTOAuthProvider, a signed client assertion could also be captured.
The advisory carries a CVSS 3.1 score of 6.5 (AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:N/A:N) for unattended providers.
In the patched versions, the client determines which issuer it expects before fetching any authorization server metadata on every path. The client refuses metadata whose issuer differs, raising an OAuthFlowError. Stored registrations are bound to their issuer; a registration bound to a different issuer is discarded, forcing the client to register again.
The advisory's prescribed remediation is to upgrade to mcp 2.2.0 or 1.30.0 or later.
ANALYSIS The flaw sits in the authentication layer that MCP clients use when connecting to third-party servers. Any deployment running affected SDK versions against untrusted or third-party MCP servers should treat the upgrade as urgent.