MCP & Protocol Testing / Scenario
Boss Battle: Ship the MCP Server
One server, one release gate, four ways it can hurt someone. Sign off only if every surface is covered: contract, security, versioning, and lifecycle.
- Difficulty
- Boss
- Format
- Scenario
- Points
- 300
- Estimate
- 20 min
// MISSION BRIEF
Your Mission
You are the release gatekeeper for payments-mcp, a server three internal agents depend on. The team wants to ship today. You have the tool schema, a security finding, a version diff, and the CI config.
This is the boss: no single 'gotcha'. Judge the whole release the way a senior QA lead would, weigh each surface and make the call. Decisions are weighted; the security and contract calls carry the most points.
// FIRST CONTACT
Battle teaser
First artifact
payments-mcp: refund_payment tool
The security finding is the headline risk. What is it, and what is the fix?
- AA JSON-RPC parse error; fix the parser
- BIndirect prompt injection via a tool result (customer notes); fix by treating tool output as untrusted data and requiring explicit user confirmation for refund_payment (a destructive action)
- CA version mismatch; bump protocolVersion
- DNothing exploitable; the notes are just data
// SKILL TAGS