{
  "lesson": "28-mcp-tool-contracts-and-content",
  "title": "MCP Tool Contracts and Content",
  "questions": [
    {
      "stage": "pre",
      "question": "A tool publishes an outputSchema with type array. Which structuredContent value is valid?",
      "options": [
        "Any JSON value because outputSchema is only documentation",
        "Only an object because MCP structured results are object-only",
        "An array whose items satisfy the published schema",
        "A text block without structuredContent"
      ],
      "correct": 2,
      "explanation": "structuredContent may be any JSON value, including an array, but it must conform to outputSchema when the tool publishes one."
    },
    {
      "stage": "check",
      "question": "A tools/list result contains nextCursor as an empty string. What should the client do?",
      "options": [
        "Treat the empty string as the end of pagination",
        "Send another list request with the exact empty string cursor",
        "Replace it with cursor zero",
        "Decode it to discover the next page number"
      ],
      "correct": 1,
      "explanation": "Cursor values are opaque. An empty string is a valid non-null cursor and must be sent back exactly."
    },
    {
      "stage": "check",
      "question": "Which parameter should be rejected by a strict deployment policy when marked with x-mcp-header?",
      "options": [
        "A string token field mirrored as Token",
        "An integer shard field mirrored as Shard",
        "A string region field mirrored as Region",
        "A boolean dry_run field mirrored as Dry-Run"
      ],
      "correct": 0,
      "explanation": "The protocol advises against mirroring sensitive values. A strict client admission policy should reject token, password, secret, credential, and PII fields."
    },
    {
      "stage": "check",
      "question": "What is the main difference between a resource_link block and an embedded resource block?",
      "options": [
        "A resource link points to data fetched later, while an embedded resource carries data now",
        "A resource link is automatically authorized forever",
        "An embedded resource must also appear in resources/list",
        "Only resource links can include a URI"
      ],
      "correct": 0,
      "explanation": "A link defers retrieval and needs authorization when followed. An embedded resource places the data in the current result."
    },
    {
      "stage": "post",
      "question": "How should a client handle one malformed tool descriptor in a tools/list page?",
      "options": [
        "Reject that tool while continuing to evaluate unrelated tools",
        "Silently repair the descriptor",
        "Expose it but add a warning to the model prompt",
        "Treat every tool execution as a JSON-RPC notification"
      ],
      "correct": 0,
      "explanation": "Descriptor admission is per tool. One malformed definition should be excluded without making valid tools unavailable."
    },
    {
      "stage": "post",
      "question": "Which failure belongs in a JSON-RPC error rather than a tool result with isError true?",
      "options": [
        "A report source is temporarily unavailable",
        "A downstream API returns an actionable tool-specific error",
        "A business rule rejects a requested date",
        "The requested tool name does not exist"
      ],
      "correct": 3,
      "explanation": "An unknown tool prevents protocol dispatch and belongs in a JSON-RPC error. Execution failures that a model can act on use isError true."
    }
  ]
}