Tool Design & MCP Integration Quiz - Certifications Claude Architect Foundations

Free practice quiz: Tool Design & MCP Integration (Certifications Claude Architect Foundations). 56 multiple-choice questions with answers and explanations. Play online, no sign-up needed.

This Certifications Claude Architect Foundations practice quiz has 56 multiple-choice questions on Tool Design & MCP Integration, and 56 of them come with an explanation. Try the 5 sample questions below (tap an option to check yourself), then play the full quiz online. No sign-up needed.

Sample questions

1. What is the most likely root cause of the extra get_customer call?

   1. The alphabetical tool order leads the model to pick the first plausible match.
   2. The simpler input schema leads the model to prefer get_customer over lookup_order.
   3. The system prompt instruction to verify customers creates a keyword association.
   4. The overlapping tool descriptions leave the model unable to distinguish the tools.
   Show answer

   Answer: D. The overlapping tool descriptions leave the model unable to distinguish the tools.

   Tool descriptions are the primary mechanism the model uses for tool selection. When two tools have near-identical descriptions, the model cannot tell when each should be used and may pick the wrong one. The fix is to clearly differentiate their purpose, inputs, outputs, and boundaries.

2. What is the best fix to the code_analysis tool's error responses?

   1. Remove the isError true flag for invalid path errors so the agent treats them as valid empty results from the current file query path in the tool output.
   2. Instruct the user to stop the current session immediately and manually fix the file path each time either error appears in the tool output.
   3. Return both errors as JSON-RPC protocol errors so the MCP client can decide whether to retry the affected request or reject it outright.
   4. Add errorCategory validation with isRetryable false for invalid paths, and errorCategory transient with isRetryable true for service unavailable.
   Show answer

   Answer: D. Add errorCategory validation with isRetryable false for invalid paths, and errorCategory transient with isRetryable true for service unavailable.

   Validation errors are non-retryable and should be clearly marked as such, while service unavailable is transient and retryable. Structured error metadata lets the agent avoid wasted retries and self-correct retryable cases. Protocol errors and suppressing isError both remove information the model needs.

3. Which configuration best guarantees a tool call while preserving the model's ability to choose the correct extraction tool?

   1. Force tool_choice to {"type": "tool", "name": "extract_invoice"} on the first turn.
   2. Add a system prompt that says, "Never return conversational text; always call an extraction tool first."
   3. Set tool_choice: "any".
   4. Add strict: true to the three extraction tool definitions so any tool call matches the JSON schema exactly.
   Show answer

   Answer: C. Set tool_choice: "any".

   tool_choice any requires the model to emit a tool call but does not force a particular tool, preserving extraction-type selection. strict improves schema conformance only after the model chooses to call a tool, while prompting remains nondeterministic and forcing extract_invoice misclassifies non-invoices.

4. The Sourcegraph search tool is registered as search_code with the description 'Search code.' What is the best change to get the agent to use it effectively?

   1. Add a system prompt instruction to the agent telling it to always prefer the search_code tool over built-in Grep.
   2. Improve the search_code description to state indexed cross-repo scope, output, and preference over Grep.
   3. Rename search_code to grep_code_search so the agent treats that tool as equivalent to the Grep tool.
   4. Disable the built-in Grep tool in the agent configuration so that search_code remains the only search option.
   Show answer

   Answer: B. Improve the search_code description to state indexed cross-repo scope, output, and preference over Grep.

   MCP tool descriptions are a primary decision signal for the agent. A detailed description that states capabilities, outputs, and when to prefer the tool prevents the agent from defaulting to built-in Grep. Disabling Grep is overly blunt, and a system prompt or rename does not address the selection logic as directly.

5. Which change is most likely to make the agent use the MCP search tool when it is appropriate?

   1. Remove Grep from the allowed built-in tools in this project so the agent cannot choose it.
   2. Rewrite the tool description to explain what it searches and what output to expect.
   3. Increase the MCP output token limit so search results appear more complete than Grep results.
   4. Move the server from the project-level .mcp.json into user-level ~/.claude.json so it loads earlier.
   Show answer

   Answer: B. Rewrite the tool description to explain what it searches and what output to expect.

   Enhancing an MCP tool's description to explain its capabilities and outputs in detail helps the agent choose it over built-in tools such as Grep when the MCP tool is more appropriate.

More questions in this quiz

Answers and explanations are in the full quiz.

1. Which change would most directly address the repeated shape and task-mixing failures?
2. What is the best first step to fix this routing error?
3. What is the most effective first change?
4. Which tool description update is best?
5. Which tool interface change best reduces this misrouting?
6. What is the most effective first step to reduce invalid lookup_order calls?
7. What is the best fix for this routing error?
8. Which tool definition update best prevents this selection error?
9. What is the most effective error handling design for the web_search subagent?
10. What structured response design should the MCP tool use to let the agent make appropriate retry decisions?
11. What is the best error-handling design for document_parser?
12. What is the most likely root cause of the agent's inability to make appropriate recovery decisions?
13. What is the most effective fix?
14. What is the best tool change?
15. What is the best fix?
16. What is the best change to the tool's error response so the agent stops wasting retries and reports the violation appropriately?
17. What is the best error handling design for the subagent?
18. What is the best error response design for the extraction tool?
19. What is the best fix to the tool's response?
20. What configuration change is the best way to guarantee the first turn calls get_customer?
21. What is the best tool configuration for the review agent?
22. Which tool configuration is the best structural fix to stop the analysis subagent from making code changes?
23. What is the most likely cause of this tool spillover?
24. Which structural change is the best way to keep documentation updates flowing while preventing writes outside docs/ and admin endpoint calls?
25. Which change best addresses the context exhaustion while keeping the subagent scoped to its research role?
26. What is the most likely reason for the drop in tool-selection accuracy?
27. Which explanation best accounts for the synthesis agent's repeated web-search calls?
28. Which tool_choice configuration best guarantees extract_metadata is called first while still allowing the model to choose the appropriate next tool afterward?
29. Which principle best explains why the review subagent edits source files despite the explicit instruction not to?
30. Which tool_choice configuration best guarantees plan_research is called first while preserving normal follow-up tool choice?
31. Which setup should the team commit to the repository?
32. What is the best way to configure the experimental server so it is available only to this researcher?
33. What is the most effective integration decision for the standard Jira integration?
34. What is the best change to reduce these exploratory tool calls?
35. During a support call, the agent called get_customer and then asked the customer for the order number instead of calling lookup_order. The developer sees no lookup_order call in traces and suspects the orders server's tools were not registered. What is the most likely situation and best next step?
36. What is the best way to configure the server so it can read the token without committing a secret?
37. Why does Claude Code use the project-scoped definition instead of the user-scoped one?
38. What is the best explanation and fix?
39. Which MCP mechanism should the server expose to provide this content catalog to Claude Code?
40. Where should the developer configure this personal/experimental server?
41. What is the best change to keep the prototype available to the developer while removing the failed connection for teammates?
42. What is the best change to make to the Grep invocation?
43. Which approach best reflects the recommended incremental codebase understanding strategy?
44. What should Claude do to search for the literal text?
45. What is the best way for the agent to make a targeted edit to the validatePayload function in this large file?
46. Which tool and search strategy is best for finding all files that reference formatDate in their contents?
47. What is the most reliable sequence for tracing all consumers of the exported components?
48. What is the best correction?
49. What is the most reliable next step?
50. Which tool choice is most appropriate?
51. What is the most reliable next step to change only the target block?

More quizzes

• Agentic Architecture & Orchestration▶ Play
• Claude Code Configuration & Workflows▶ Play
• Prompt Engineering & Structured Output▶ Play

Browse all Claude Architect Foundations quizzes

More to play