Summary
A critical authentication bypass in Google MCP Toolbox causes opaque tokens to be accepted when an OAuth 2.0 introspection response omits the required active field. An attacker may gain access to protected tools and connected data sources when the toolbox trusts an incomplete or malformed introspection response.
Affected context
The issue occurs in deployments using the generic opaque-token validation path and an OAuth 2.0 introspection endpoint compliant with RFC 7662. Risk is present when the endpoint can return a response that omits the mandatory active property and the toolbox treats the resulting absent value as acceptable rather than invalid.
Exploit preconditions
- Attack surface: Opaque bearer-token validation through an OAuth 2.0 introspection endpoint
- Required access: A token or request reaching a protected MCP Toolbox interface, plus an introspection response path that can produce incomplete validation data
- Trigger point: Processing the introspection response in
validateOpaqueToken - Why exploitation is possible: The implementation rejects an explicitly false
activevalue but does not reject a missing value, allowing an incomplete response to be interpreted as successful validation
Why this happens
The underlying bug is fail-open authentication logic combined with ambiguous handling of a nullable boolean. The introspection response maps active to a pointer, so an omitted field becomes nil; the validation condition only denies access when the pointer is non-nil and false. This maps to CWE-287: Improper Authentication, because the implementation does not reliably establish that the presented token is active before authorizing access.
Potential impact
- Unauthorized users may access protected MCP tools
- Attackers could reach connected data sources without valid active tokens
- Sensitive data may be disclosed or modified
- Downstream services may be invoked under an incorrectly authenticated identity
- The affected deployment could lose confidence in token-based access controls
General mitigation guidance
Mitigation generally focuses on making token validation fail closed and treating incomplete introspection responses as invalid.
Common defensive patterns include:
- Require the introspection response to contain a valid boolean
activefield - Reject missing, null, malformed, or unexpected authentication attributes
- Apply strict response schemas before authorization decisions
- Keep token validation and authorization checks separate and explicit
- Limit the privileges available to MCP tools and connected data sources
- Use secure defaults when upstream identity services are unavailable or return ambiguous data
- Protect introspection communications and enforce trusted endpoint configuration
How teams detect issues like this
Organizations can look for authentication checks that treat absent or nullable identity fields as successful validation.
Useful approaches include:
- Static analysis for conditional checks that reject only explicit false values
- Code review of pointer, nullable, and optional fields in authentication responses
- Tests covering missing, null, malformed, and contradictory introspection attributes
- Contract validation for RFC 7662 responses before authorization
- Runtime monitoring for successful requests associated with incomplete introspection responses
- Policy enforcement requiring fail-closed behavior in token validation components
Automated code review and security testing tools may help identify these patterns during development.

