Implementation guide
OWASP AISVS v1.01
Model Context Protocol (MCP) Security · Schema, Message, and Input Validation
Drop tools with invalid header annotations
Verify that MCP clients reject tool definitions containing invalid mirrored-header annotations, excluding the affected tool rather than the entire tool list.
OWASP AISVS v1.01, requirement 10.4.11
In plain English
If one tool definition arrives with invalid mirrored-header annotations, the client should drop that tool and keep the rest of the list. Rejecting everything would let one bad tool disable a whole server, and teams respond to that by switching validation off.
What good looks like
- The client validates the mirrored-header annotations on each tool definition.
- A tool with invalid annotations is excluded and cannot be called.
- The other tools in the list stay available.
- The exclusion is logged with the reason.
How teams usually do it
- Per-tool validation in the client, with errors handled tool by tool.
- Surfacing excluded tools to the operator so the server owner can fix them.
Suggested evidence
- A test with one invalid tool definition in a list, showing only that tool is dropped.
- Log entries for the exclusion.
Common mistakes
- Rejecting the whole list.
- Accepting the tool and ignoring the bad annotation, which leaves gateways enforcing policy on headers that mean something else.
Not sure which requirements apply to your system? Take the assessment, or browse every guide in this chapter.
Requirement text from the OWASP AI Security Verification Standard, licensed under CC BY-SA 4.0. This guidance is shared under the same licence.