CCARF Practice Question

Paper 1 · Question 2

D2 Tool Design & MCP Integration · Paper 1

A developer productivity platform has a tool called analyze_content with the description 'Analyses content from various sources.' Logs show the agent uses it for web scraping, document parsing, and code analysis indiscriminately, leading to poor results for each use case. What is the most effective fix?

• A.Add detailed few-shot examples to the system prompt showing which types of content the tool handles well and which it does not.
• B.Improve the tool description to list all supported content types and their expected formats.
• C.Split analyze_content into purpose-specific tools: extract_web_results, parse_document, and analyze_code, each with a narrow remit.
• D.Implement a routing layer that inspects the input content type and dispatches to the appropriate internal processing function within analyze_content.

Show answer & explanations

Correct answer: C

A: Few-shot examples add token overhead and treat the symptom. The root cause is that a single generic tool is trying to serve three distinct use cases, each requiring different processing logic.

B: A better description on a generic tool does not change the fact that the tool's processing logic is too broad. When a tool serves multiple distinct purposes, the fix is to split it into purpose-specific tools.

C: Tool splitting replaces a generic tool with purpose-specific alternatives that each have a clear, narrow description and defined input/output contracts. Each resulting tool handles one use case well, which eliminates ambiguity about when to use it.

D: An internal routing layer hides the content type distinction from the agent. Splitting into separate tools exposes the distinction through tool descriptions, enabling the LLM's natural language understanding to select the right tool.

Related Tool Design & MCP Integration questions

More: Paper 1 · All practice questions