CCAR-F Tool Design and MCP Integration Practice Question
During an MCP tool invocation, a user asks Claude to look up an order by its ID. The MCP server's tool handler returns a successful result, but the response payload is 12 MB of raw JSON containing nested line items, shipping events, and audit metadata. Claude's reply to the user is vague and omits key order details. Which change to the MCP tool design best addresses this?
⚠ Common exam trap
The trap here is assuming a successful tool call means the tool design is correct, when an oversized or unfocused result payload can still degrade the answer.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Return only a compact, task-relevant projection of the order data from the MCP server, with pagination or a detail-fetch tool for anything further.
The tool returned successfully, so transport, timeouts, and output token limits are not the issue. The failure is context quality: a 12 MB nested payload buries the fields the user cares about and degrades the model's answer. Shrinking the tool result to a task-relevant projection, and exposing deeper data through pagination or a dedicated detail tool, restores precision without losing capability.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Raise the MCP server's request timeout so the 12 MB response can be fully streamed before Claude processes it.
Why it's wrong here
Timeout and streaming behavior affect whether the transfer completes, not whether the model can use the content. The scenario states the tool already returns successfully, so latency and truncation are not the failure mode. Extending timeouts would add delay without improving answer quality, and could mask genuine transport problems in production.
- ✓
Return only a compact, task-relevant projection of the order data from the MCP server, with pagination or a detail-fetch tool for anything further.
Why this is correct
Large payloads dilute the model's attention and crowd out the user's actual question. Returning a small, purpose-built projection such as order status, total, and estimated delivery keeps the signal high, and offering a separate detail tool or pagination lets the model pull deeper data only when the conversation requires it. This is the standard MCP tool-design pattern for verbose upstream systems.
- ✗
Add a system prompt instruction telling Claude to always ignore audit metadata and shipping events when answering order questions.
Why it's wrong here
Prompt instructions cannot reliably compensate for a tool result that floods the context with irrelevant structure. The model still has to ingest the full payload, and instructions about what to ignore are far weaker than simply not sending the noise. Fixing this at the prompt layer rather than the tool contract also forces every future consumer of the tool to repeat the workaround.
- ✗
Increase the max_tokens setting on the Claude request so the model has more room to summarize the 12 MB payload.
Why it's wrong here
max_tokens controls output generation length, not how much input is retained or attended to. Raising it does not make the model extract the relevant fields from an oversized payload, and the input may exceed the context window entirely. The problem is the shape and size of the tool result, not the response budget, so this leaves the root cause untouched.
About these practice questions
One of 271 original CCAR-F practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official Anthropic exam blueprint
This CCAR-F practice question is part of Courseiva's free Anthropic certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the CCAR-F exam.