Free CCAR-P Integration Practice Questions

The Integration domain makes up 19% of the Anthropic CCAR-P exam. Practice free questions covering enterprise system integration, Claude API patterns, MCP servers and clients, data pipelines, authentication and authorization boundaries — each with the correct answer and a detailed explanation. The real exam mixes multiple-choice and multiple-response items; these practice questions use a single-best-answer format.

CCAR-P Integration Sample Questions with Answers

Sample Question 1 — Integration

A telecommunications provider is deploying a Claude-based service agent that reads account history and issues refunds. Routine refunds below $50 may be automated, but larger refunds require supervisor approval. The prototype uses one account-management tool that can update plans, change addresses, close accounts, and issue refunds of any amount. All calls are logged, and the team proposes instructing Claude to use only the refund operation. Which change BEST addresses the production risk while preserving routine automation?

  1. A. Retain the account-management tool, add anomaly alerts, and review unusual actions after each shift.
  2. B. Replace it with read-only account access and a narrow refund tool that enforces amount limits and approval tokens. (Correct answer)
  3. C. Retain the account-management tool, strengthen its system instructions, and require Claude to explain each action.
  4. D. Send every proposed account action to a supervisor before allowing any tool invocation to execute.

Correct answer: B

Explanation: Correct answer (B): The narrow refund tool is best because the agent needs automated access only to low-value refunds, while larger refunds require approval. Enforcing operation and amount boundaries in trusted code applies least privilege and keeps authorization outside the model. This prevents unauthorized actions rather than merely detecting them, while avoiding unnecessary review of permitted routine refunds. Why the other options are wrong: - Option A: An architect might add alerts for accountability, but post-action detection does not remove the agent's ability to close accounts or issue excessive refunds. - Option C: Stronger instructions and explanations may improve behavior and auditability, but they are not deterministic controls over the broad administrative capability. - Option D: Universal approval would control risk, but the scenario explicitly permits automated refunds below $50, making this operationally excessive.

Sample Question 2 — Integration

An online marketplace classifies 120,000 support tickets daily. A lower-cost Claude configuration meets the 1.5-second SLA and achieves 96% overall routing accuracy, but reaches only 88% on the 3% of tickets involving account takeover. A more capable configuration reaches 98% on that segment but doubles average latency and triples total inference cost. The business requires at least 97% accuracy for account-takeover tickets while holding total cost near the current budget. What should the architect recommend?

  1. A. Use the more capable configuration for every ticket and offset its cost by reducing operational logging.
  2. B. Keep the lower-cost configuration for every ticket and add a second identical classification call when confidence is low.
  3. C. Route likely account-takeover or ambiguous tickets to the more capable configuration and evaluate routing misses separately. (Correct answer)
  4. D. Run both configurations for every ticket and use a third model to resolve classifications on which they disagree.

Correct answer: C

Explanation: Correct answer (C): Selective escalation concentrates additional capability and cost on the small segment with the stringent quality requirement. The architecture must also evaluate missed high-risk cases because routing quality governs whether escalation works. This risk-based segmentation balances correctness, latency, and cost more effectively than applying expensive processing to the entire workload. Why the other options are wrong: - Option A: A universal upgrade would improve the high-risk segment, but it violates the stated cost and latency constraints, while reducing logging weakens operations. - Option B: A repeated call may help with stochastic errors, but no evidence shows that the lower-cost configuration can reach the required high-risk accuracy. - Option D: Ensembling is technically viable for difficult cases, but three calls on every ticket add unnecessary latency and cost to the low-risk majority.

Sample Question 3 — Integration

A cloud operations assistant supports 4,000 incident investigations per day by searching runbooks, calling monitoring APIs, and asking Claude to propose remediation steps. Ten percent of requests intermittently exceed the 8-second SLA, but application logs cannot show whether retrieval, model calls, or tools caused the delay. Full transcripts contain customer identifiers and cannot be retained broadly. Which observability change would BEST support diagnosis?

  1. A. Retain all transcripts in encrypted storage and have operators manually compare slow and fast investigations.
  2. B. Track only average end-to-end latency and alert when the daily average exceeds eight seconds.
  3. C. Add more retries to every monitoring API call and record the final success status in application logs.
  4. D. Propagate trace IDs across retrieval, model, and tool calls, recording structured latency, token, and failure fields with privacy-aware sampling. (Correct answer)

Correct answer: D

Explanation: Correct answer (D): Distributed trace correlation identifies which integration stage causes tail latency while structured fields support fleet-wide aggregation. Privacy-aware sampling addresses the restriction on transcript retention. This follows the production principle that model, retrieval, and tool behavior must be observable as separate but correlated components rather than diagnosed from averages or manual transcript review. Why the other options are wrong: - Option A: Transcripts can help inspect individual cases, but broad retention conflicts with the privacy constraint and does not efficiently isolate component latency at scale. - Option B: Average latency is inexpensive to monitor, but it can hide tail behavior and cannot identify the responsible integration layer. - Option C: Retries may mitigate transient failures, but indiscriminate retries can increase latency and do not diagnose the measured bottleneck.

Sample Question 4 — Integration

A pharmaceutical company indexes research reports, public procedures, and acquisition documents in one hybrid search service. Access varies by employee, project, and document classification. The assistant must provide traceable citations, and restricted text must never enter an unauthorized employee's model context. The prototype retrieves globally, reranks results, and asks Claude to omit passages whose metadata does not match the user's role. Which design BEST satisfies the requirement?

  1. A. Retrieve globally, redact restricted passages after generation, and retain citations only for the passages shown to the user.
  2. B. Preserve access metadata during ingestion, security-trim every retrieval and cache layer by user authorization, and pass only authorized chunks with provenance. (Correct answer)
  3. C. Create separate indexes for each classification level, let the application select one index, and rely on Claude to enforce project membership.
  4. D. Encrypt restricted chunks in the shared index, give the retrieval service a decryption key, and ask Claude not to cite restricted sources.

Correct answer: B

Explanation: Correct answer (B): Security trimming must occur before restricted content enters model context and must cover semantic search, lexical search, reranking, and caching. Preserving access metadata and provenance supports both authorization and traceable citations. This structural boundary prevents disclosure rather than relying on the model or an output filter after unauthorized retrieval has already occurred. Why the other options are wrong: - Option A: Output redaction can be a compensating control, but restricted content has already entered context and may influence or leak through the answer. - Option C: Index partitioning can support classification boundaries, but delegating project-level authorization to Claude leaves a required permission dimension unenforced. - Option D: Encryption protects data at rest, but a retrieval service holding the key can still expose decrypted restricted chunks to unauthorized model requests.

Sample Question 5 — Integration

A bank's policy assistant cites a two-month-old travel policy even though the source repository was updated yesterday. Retrieval traces show only the older document version, and the prompt correctly requires answers to use retrieved sources. Compliance needs the corrected policy available by the next business day. What should the architect investigate FIRST?

  1. A. Verify ingestion, version metadata, index refresh, and retrieval-cache invalidation for the updated policy. (Correct answer)
  2. B. Strengthen the prompt by telling Claude that recent policies should take precedence over older versions.
  3. C. Move the assistant to a more capable model that can better identify outdated policy language.
  4. D. Add an output evaluator that flags answers containing dates older than the current calendar month.

Correct answer: A

Explanation: Correct answer (A): The trace proves that the updated policy never reached model context, so the first action is to inspect the freshness path from ingestion through cache invalidation. Claude cannot ground an answer in a source it did not receive. Fixing the stale retrieval layer removes the root cause and directly meets the compliance deadline. Why the other options are wrong: - Option B: Recency instructions can help when multiple versions are present, but the trace shows that only the older version was retrieved. - Option C: A stronger model may reason better over available context, but it cannot use the missing updated policy. - Option D: An evaluator could detect some stale answers, but it would not refresh the index and may misclassify valid older policies.

Sample Question 6 — Integration

An enterprise operations assistant can access 240 tools across finance, facilities, identity, and cloud platforms. Each request typically needs no more than three tools, and users have different permissions. Supplying every tool definition consumes substantial context and has increased incorrect tool selection. The catalog changes several times per week, while the interactive SLA is four seconds. Which design should the architect choose?

  1. A. Keep all tool definitions in context, shorten each description, and instruct Claude to ignore tools outside the user's department.
  2. B. Divide the catalog into four static prompts by department and grant each department access to every tool in its prompt.
  3. C. Expose an authorization-aware discovery surface, then load only relevant tool definitions and measure the added discovery latency. (Correct answer)
  4. D. Train a separate model for each platform and permanently embed that platform's complete tool catalog in its system prompt.

Correct answer: C

Explanation: Correct answer (C): Progressive discovery reduces context consumption and selection ambiguity because each request needs only a few of 240 tools. Filtering discovery by authorization prevents unavailable capabilities from being exposed, while measuring discovery latency protects the four-second SLA. The design accepts added routing complexity only where the catalog's scale and rate of change justify it. Why the other options are wrong: - Option A: Shorter descriptions may reduce context use, but all 240 tools remain exposed and prompt instructions do not enforce user authorization. - Option B: Static grouping can reduce catalog size, but department-level grants are too coarse for the stated user-specific permissions and changing catalog. - Option D: Platform-specific models could narrow selection, but training and maintaining multiple models is unnecessary and still exposes complete platform catalogs.

Sample Question 7 — Integration

A distributor's assistant answers conceptual questions from product manuals and exact questions about contract IDs, current inventory, and product codes. Semantic retrieval performs well on troubleshooting questions but often confuses similar identifiers. The document index refreshes nightly, while inventory changes every minute. Users require citations identifying whether evidence came from a document or the system of record. Which retrieval architecture BEST fits these query patterns?

  1. A. Route exact live-data queries to authorized structured services, use hybrid retrieval and reranking for documents, and preserve source provenance. (Correct answer)
  2. B. Embed manuals and nightly inventory exports together, increase semantic top-k, and ask Claude to choose the newest matching chunk.
  3. C. Use keyword search for every query, return all exact token matches, and let Claude infer conceptual relationships from those results.
  4. D. Use semantic retrieval for every query, add identifier examples to the prompt, and refresh all embeddings every fifteen minutes.

Correct answer: A

Explanation: Correct answer (A): Exact identifiers and minute-by-minute inventory values belong in authorized structured lookups, while conceptual document questions benefit from hybrid lexical and semantic retrieval with reranking. Preserving provenance satisfies the citation requirement. Matching retrieval strategy to data shape and freshness avoids forcing one mechanism to handle fundamentally different workloads. Why the other options are wrong: - Option B: A combined index is simple, but nightly inventory exports cannot satisfy minute-level freshness, and larger top-k does not ensure exact identifier matching. - Option C: Keyword search handles codes well, but using it exclusively reduces recall for conceptual troubleshooting questions that semantic retrieval already handles effectively. - Option D: Prompt examples may improve formatting, but semantic search remains unreliable for similar exact IDs, and periodic embedding refreshes still lag live inventory.

Sample Question 8 — Integration

A market-intelligence assistant searches three independent news providers, extracts company names, then queries regulatory databases using the discovered names. Distributed traces show that the three news searches run sequentially and consume six of the workflow's nine seconds. Regulatory searches cannot begin until extraction completes. The SLA is six seconds, and speculative regulatory queries would create substantial vendor charges. Which change MOST directly improves latency without violating the workflow dependencies?

  1. A. Run every news and regulatory query concurrently, then discard results for companies that extraction does not identify.
  2. B. Run the three news searches concurrently, preserve extraction before regulatory lookup, and cancel unnecessary branches when sufficient evidence exists. (Correct answer)
  3. C. Keep execution sequential, increase each tool timeout, and add retries so slow providers have more opportunity to respond.
  4. D. Replace the workflow with multiple autonomous agents that independently choose providers and exchange findings until they agree.

Correct answer: B

Explanation: Correct answer (B): The trace identifies three independent searches as the measured bottleneck, so they can run concurrently without changing semantics. Regulatory lookup must remain after entity extraction because it has a genuine data dependency, and cancellation limits unnecessary cost. This is dependency-aware optimization rather than indiscriminate parallelism. Why the other options are wrong: - Option A: Full speculative parallelism could reduce wall-clock time, but regulatory queries lack required company names and would generate the substantial charges the scenario prohibits. - Option C: Longer timeouts and retries may improve completion rates, but they increase latency and do not address the sequential independent searches shown by traces. - Option D: Multiple agents may support genuinely independent analysis, but they add coordination, calls, and nondeterminism to a workflow with known dependencies.

Keep Practicing

Take the 10-question CCAR-P quick-start test across all 7 domains, or return to the CCAR-P practice test hub for 200+ questions, exam details, and a study plan. Drill the other domains: Integration, Solution Design & Architecture, Evaluation, Testing & Optimization, Governance, Safety & Risk Management, Stakeholder Communication & Lifecycle Management, Claude Models, Prompting & Context Engineering, and Developer Productivity & Operational Enablement.