Technical impact
The Model Context Protocol Python SDK performs server‑side bearer authentication using a TokenVerifier that does not independently validate the token’s audience (aud) claim when the authorization server issues tokens for multiple resource servers. As a result, a bearer token issued for one MCP server can be accepted by another server, allowing the holder to authenticate to an unintended resource and exercise the permissions encoded in that token. This flaw effectively bypasses intended audience constraints, enabling an attacker who holds a valid token for one server to gain unauthorized access to a different MCP server, potentially exposing sensitive data or operations governed by that resource.
Affected Systems
The vulnerability affects Model Context Protocol Python SDK releases from version 1.10.0 up to (but not including) 1.30.0, as well as 2.x releases prior to 2.2.0. The fix was introduced in SDK 1.30.0 and 2.2.0. Any deployments using these earlier versions are exposed.
Risk and Exploitability
The CVSS score of 6.8 indicates a serious severity, and the EPSS score of less than 1% suggests a very low probability of exploitation at the time of analysis. The vulnerability is not listed in CISA’s KEV catalog. The likely attack vector is a simple token replay across resource servers: an adversary with a valid token for one resource can reuse it against another MCP server because the SDK does not enforce the audience claim, making the exploit trivial for whoever can obtain or intercept a token.
Github GHSA