--- title: "Free CCDV-F Mock Exam, Timed, No Sign-up | Cred Farmer" url: https://credfarmer.com/practice/ccdv-f description: "53 CCDV-F questions in a free 120-minute mock, no account needed. Sit it timed, quick or untimed, then get a score by domain and a reason for every option." last_updated: 2026-10-29 publisher: "Cred Farmer, an independent study platform, not affiliated with or endorsed by Anthropic or Pearson VUE" --- No account needed # Free CCDV-F mock exam, timed, with a reason for every option 53 questions written for the Claude Certified Developer - Foundations blueprint, sat at the official pace of about 2.3 minutes an item. No sign-up, nothing saved, nothing sent anywhere. Want more? The post [CCDV-F worked scenarios](https://credfarmer.com/blog/ccdv-f-practice-questions) works 10 more items option by option, and the [CCDV-F study guide](https://credfarmer.com/certifications/claude-certified-developer-foundations#questions) shows 6 more with their keys: 69 distinct items with this paper, none behind an account. 53 items · 120 minutes official · 720 to pass · No account **The CCDV-F exam changes on 7 December 2026.** Pearson VUE says the CCDV-F exam includes updated content from 7 December 2026, and that the updated exam guide is published on 17 November 2026. This page follows Exam Guide v1.0 (July 2026), the guide to prepare from if you sit before 7 December 2026. If you sit on or after that date, prepare from the updated guide. We will compare it with v1.0 when it is published and update this page. [Pearson VUE's notice](https://www.pearsonvue.com/us/en/anthropic.html), read on 29 October 2026. **53**Items in this mock **120**Official minutes **720**Scaled score to pass **$125**Official exam fee **8**Official domains Question bank dated 22 October 2026. Checked against Exam Guide v1.0 (July 2026) on 17 September 2026. Three ways to sit it ## Start the CCDV-F mock Answer in this tab and get a raw score by domain with the key and the reason each option fails. Every mode scores the same way and shows the same review afterwards. The clock is the only difference, and it is the difference most candidates underestimate. Or work through the paper below one question at a time, checking each as you go. ### Full, timed All 53 items in 120 minutes, about 2.3 minutes an item, the official pace. The sitting submits itself when the clock reaches zero. ### Quick, timed 30 items drawn evenly across the 8 domains, in 68 minutes at the same pace. A different draw each time. ### Untimed review Every item with no clock. Flag what you are unsure of, submit when you are done, and read the reasons. ## Filter the paper by domain - [All domains 53](https://credfarmer.com/practice/ccdv-f) - [D1 Agents and Workflows 8](https://credfarmer.com/practice/ccdv-f?domain=1) - [D2 Applications and Integration 17](https://credfarmer.com/practice/ccdv-f?domain=2) - [D3 Claude Code 2](https://credfarmer.com/practice/ccdv-f?domain=3) - [D4 Eval, Testing, and Debugging 1](https://credfarmer.com/practice/ccdv-f?domain=4) - [D5 Model Selection and Optimization 9](https://credfarmer.com/practice/ccdv-f?domain=5) - [D6 Prompt and Context Engineering 6](https://credfarmer.com/practice/ccdv-f?domain=6) - [D7 Security and Safety 4](https://credfarmer.com/practice/ccdv-f?domain=7) - [D8 Tools and MCPs 6](https://credfarmer.com/practice/ccdv-f?domain=8) Answer as you read ## The CCDV-F paper: every question 53 items, each tagged with its official domain. Work through the paper one question at a time: pick an option to check it, or ask for the key, and read why each option holds or fails. Nothing is saved. For a scored sitting against the clock, use the start modes above. Checked 0 of 53, 0 correct [Sit it timed instead](https://credfarmer.com/practice/ccdv-f#pp-start-heading) 1. Question 1 · Domain 2: Applications and Integration · Single answer A product owner sets two requirements for a mortgage assistant: answers start appearing within three seconds, and every quoted rate matches a rate table that changes several times a day. Which design meets both? Select the best answer. - A. Fetch the current rate with a tool call at request time, then stream the answer. - B. Put the rate table in a cached system prompt and refresh it every night. - C. Queue each question to the Message Batches API and recheck the rate first. - D. Use the largest model with extended thinking. 2. Question 2 · Domain 1: Agents and Workflows · Single answer A team processes customs declarations in four fixed steps: extract fields, classify the goods, check them against a tariff table, and file. Auditors must see which step produced each value. Which architecture fits best? Select the best answer. - A. A workflow that chains one Claude call per step, with code checks between steps. - B. One autonomous agent that picks the order of the four steps itself. - C. A manager agent that hands each declaration to a subagent that picks its own tools. - D. A single prompt that performs all four steps in one response. 3. Question 3 · Domain 2: Applications and Integration · Single answer An IDE plugin shows each 600-token code explanation all at once, after about eight seconds of an empty panel. Many developers close it first. Explanations must stay complete. What should you change? Select the best answer. - A. Move the explanations to a smaller, faster model tier. - B. Store each finished explanation and reuse it when the same code is opened again. - C. Generate explanations for every file in the background when a project opens. - D. Enable streaming and render each text delta the moment it arrives. 4. Question 4 · Domain 2: Applications and Integration · Single answer Three analysts edit a production prompt in an admin console. After a wording change broke a required disclaimer, nobody could tell who changed it or what the earlier text said. What practice prevents a repeat? Select the best answer. - A. Limit console access to one senior analyst who then makes every prompt change. - B. Put the prompt in version control, behind code review. - C. Pin the model to a dated version and keep editing the prompt in the console. - D. Have each analyst add a dated comment at the top of the prompt describing each edit. 5. Question 5 · Domain 1: Agents and Workflows · Single answer A hospital is deploying a discharge-summary agent. Compliance requires the agent loop, tool execution and patient records to stay inside the hospital's own cloud account. Which deployment model fits? Select the best answer. - A. Use an Anthropic-hosted managed agent and copy its logs into the hospital's account. - B. Self-host the agent harness in the hospital's account and call the model through the API. - C. Use an Anthropic-hosted agent and keep patient records in its session memory. - D. Run the agent in Claude Desktop on each clinician's laptop. 6. Question 6 · Domain 2: Applications and Integration · Single answer A tool's input schema declares a cause field as a free string. Over 10,000 calls it held 61 spellings of six fixed causes, and exact-match routing failed on 18 percent. Which schema change fixes it? Select the best answer. - A. Keep the free string and add fuzzy matching in the router. - B. List the six values as examples in the tool description, keeping the string type. - C. Declare the cause field as an enum of the six permitted values. - D. Split the field into six boolean flags, one per cause. 7. Question 7 · Domain 3: Claude Code · Single answer You want Claude Code to review every pull request in CI and return findings as JSON. No one is at a terminal, and the job must never edit files. How should you run it? Select the best answer. - A. In an interactive session with auto-accept on, parsing the transcript afterward. - B. In headless mode with JSON output and only read-only tools allowed. - C. Through a custom slash command that a reviewer runs from their own session. - D. Through Claude Desktop installed on the CI runner. 8. Question 8 · Domain 1: Agents and Workflows · Select 2 An agent checks 120 grants, each with a 30,000-token ledger, and its window fills by grant 40. You plan to hand each grant's check to a subagent. Which two benefits can you expect? (Select two.) Select exactly 2.**0/2 selected.** - A. Subagents automatically see the parent's history and earlier findings. - B. Each ledger stays in its subagent's own context, and only the finding comes back. - C. A fresh context for each check makes every finding identical across reruns. - D. Independent checks can run in parallel and cut the total run time. - E. Each subagent inherits the parent's instructions and the funder rules without being passed them. 9. Question 9 · Domain 2: Applications and Integration · Single answer Each request sends the changing clause first, then a fixed 38,000-token playbook, with cache_control set at the end. Cache hits stay near zero. Which arrangement produces cache hits? Select the best answer. - A. Put the playbook first, then the breakpoint, then the clause. - B. Route the same requests through the Message Batches API, keeping the order. - C. Keep the clause first and set the cache breakpoint on the clause block. - D. Compress the playbook into a 4,000-token summary and drop cache_control. 10. Question 10 · Domain 2: Applications and Integration · Single answer A prompt refined in claude.ai gave well-formatted answers that declined legal advice. Sent as the user turn of an API call, it gave plain output and sometimes legal guidance. What explains the difference? Select the best answer. - A. The API serves a different model than claude.ai. - B. The API ignores instructions placed in the user turn. - C. The API has no default system prompt; the app must now supply that context itself. - D. Claude.ai filters output after generation, and the app needs its own filter. 11. Question 11 · Domain 1: Agents and Workflows · Single answer A custom agent loop logs each tool_use block but never runs the tool. It drops the block and asks the model to answer, so replies quote figures no tool returned. What fixes the loop? Select the best answer. - A. Set tool_choice to require the tool on every request. - B. Raise max_tokens until the model can call the tool and answer in one response. - C. Run the tool, return a tool_result with the matching id, and call the model again. - D. Paste the data the tool would return into the system prompt and remove the tool definition. 12. Question 12 · Domain 2: Applications and Integration · Select 2 You will regenerate 250,000 listing descriptions with the Message Batches API. You have only used the synchronous Messages API before. Which two behaviors must the pipeline handle? (Select two.) Select exactly 2.**0/2 selected.** - A. Results can come back in any order and must be matched by custom_id. - B. Requests in one batch share a conversation and inherit earlier context. - C. Results arrive as streamed deltas that must be reassembled. - D. Batch pricing applies only above a minimum number of requests. - E. Processing may last up to 24 hours, and the pipeline must poll for status. 13. Question 13 · Domain 2: Applications and Integration · Single answer A page summarizes up to 300 independent reviews, one synchronous call after another, and takes 45 seconds to build. The rate limit is well above current use. What most directly cuts the build time? Select the best answer. - A. Concatenate all 300 reviews into a single request and ask for one combined summary. - B. Use the fastest model tier for every call and leave the loop fully sequential. - C. Start an unbounded thread for each review. - D. Make the calls concurrently with the async client, capped by a semaphore. 14. Question 14 · Domain 4: Eval, Testing, and Debugging · Single answer An integration retries every failed Claude API call at once, up to ten times. Logs show rising 429s, a malformed 400 retried ten times, and a 401 after a key rotation. Which retry policy fits? Select the best answer. - A. Retry 429 and 5xx with backoff; fix 400 and 401. - B. Retry every error type with a fixed 30-second delay between each attempt. - C. Turn off retries and fail each request on its first error. - D. Retry 400 and 401 with backoff, and fail every 429 error immediately. 15. Question 15 · Domain 1: Agents and Workflows · Single answer CLAUDE.md tells an agent to run a pricing linter after every edit to a rules file, yet the linter was skipped in 11 of 60 sessions. How can you make it run every time? Select the best answer. - A. Move the instruction to the first line of CLAUDE.md, in capital letters. - B. Register a PostToolUse hook that runs the linter whenever the rules file is edited. - C. Expose the linter as a custom tool described as required after each edit. - D. Add a reviewing subagent at the end of each session. 16. Question 16 · Domain 2: Applications and Integration · Single answer A triage system matched clerks on 93 percent of complaints in offline testing. The director wants clerks reassigned next month, but misrouted safety complaints carry legal risk. Which rollout approach fits best? Select the best answer. - A. Run it in shadow mode first, then promote queue by queue with rollback. - B. Cut over fully next month and rely on production dashboards to spot any misrouting. - C. Keep it in shadow mode with no cutover, as a suggestion tool. - D. Delay launch until offline agreement exceeds 99 percent. 17. Question 17 · Domain 2: Applications and Integration · Single answer A clinical-notes assistant keeps one conversation open per clinician for a whole shift across 20 or more patients. Some notes mention another patient's medication. Which change removes this cross-patient leakage? Select the best answer. - A. Tag each patient's turns with that patient's record number. - B. Use the model with the biggest context window. - C. Enable prompt caching on the earlier turns of the conversation. - D. Start a fresh conversation for each patient encounter. 18. Question 18 · Domain 1: Agents and Workflows · Single answer One agent reviews 400 contracts against a checklist and writes a combined risk summary. After about 30 contracts it skips items and mixes up clauses between contracts. Which change addresses this? Select the best answer. - A. Keep one agent and compact the conversation whenever it nears the context limit. - B. Upgrade the single agent to the top model tier and keep the rest of the design. - C. Write each finished review to a memory file and have the agent read it before every new contract. - D. Use a supervisor that sends each contract to its own subagent. 19. Question 19 · Domain 2: Applications and Integration · Single answer Policy requires all model inference to run through the cloud provider where your company holds its contracts and audit logging. The prototype uses Anthropic's first-party SDK. What should the team expect when moving? Select the best answer. - A. Platform-specific Claude variants that need every prompt and eval rebuilt. - B. No code or credential changes; the SDK finds the provider from the key. - C. The same Claude models, with a new endpoint, client setup and credentials; prompts carry over. - D. Access through batch jobs only, with no interactive requests. 20. Question 20 · Domain 2: Applications and Integration · Single answer Claude Code migrated 200 of 900 files to async/await in one session and one commit. Tests then showed 31 failures, and tracing each cause took a week. How should you migrate the remaining 700 files? Select the best answer. - A. Migrate them all in one session on the highest model tier. - B. Have the model edit the failing tests until they pass. - C. Migrate small batches of related files, running tests and committing after each. - D. Disable the failing tests until the migration ends, then fix them all in one pass. 21. Question 21 · Domain 2: Applications and Integration · Single answer A claims service names its model with an alias that follows the newest release. After a release, responses gained a sentence before the JSON and 40 percent failed parsing. Which practice prevents a repeat? Select the best answer. - A. Keep the alias and add a parser step that strips any text before the JSON. - B. Pin a dated model version and adopt new ones after the eval suite passes. - C. Replace the alias with the top model tier's alias and keep the current parser. - D. Version the prompt and leave the alias unchanged. 22. Question 22 · Domain 1: Agents and Workflows · Single answer Two engineers need an internal agent that reads a repository, edits files, runs tests and asks for approval before shell commands, with a pilot due in three weeks. Which starting point fits best? Select the best answer. - A. A hand-written loop on the Messages API with its own approval step. - B. The Claude Agent SDK, with its built-in loop and tools. - C. The Claude Agent SDK with no system prompt and no permission settings. - D. A custom loop; the SDK runs only inside the interactive Claude Code terminal. 23. Question 23 · Domain 2: Applications and Integration · Single answer A client posts each outage notice to a REST endpoint that accepts an idempotency key, and retries after a 10-second timeout. Some customers got duplicates. Which change stops them but keeps the retry? Select the best answer. - A. Generate a new key for each attempt and reconcile duplicates afterward. - B. Raise the timeout to 60 seconds and keep the request unchanged. - C. Remove the retry and log each timeout for manual follow-up. - D. Send the same idempotency key on the first request and on every retry of it. 24. Question 24 · Domain 3: Claude Code · Single answer In a monorepo with a root CLAUDE.md, the payments team needs stricter rules that apply only when Claude Code works inside services/payments. Where should those rules go? Select the best answer. - A. In a CLAUDE.md file inside the services/payments directory. - B. In the root CLAUDE.md under a heading saying they apply to payments only. - C. In each payments engineer's user-level CLAUDE.md file. - D. In the user-level settings.json of each payments engineer. 25. Question 25 · Domain 2: Applications and Integration · Single answer A support assistant sends the instructions, the customer's message and the order history as one user-turn string. Sometimes it obeys text in the customer's message as an instruction. Which change sets the missing boundary? Select the best answer. - A. Put the instructions in the system prompt and the order history in data tags. - B. Move the order history into the system prompt. - C. Add a sentence that tells the assistant to ignore any instructions in a customer's message. - D. Switch to the highest-capability model tier and keep the current one-string request. 26. Question 26 · Domain 1: Agents and Workflows · Single answer A maintenance-planning agent runs 80 or more turns. By turn 50 it re-inspects machines it already cleared, because tool output has pushed early decisions out of context. Which pattern fixes this without a larger window? Select the best answer. - A. Switch to another agentic framework that has built-in state handling. - B. Restart the session every 40 turns with only the original task description. - C. Save decisions to a memory store read at every step. - D. Copy every tool result into the system prompt and keep it there for the whole run. 27. Question 27 · Domain 2: Applications and Integration · Single answer A formatter hook runs for three of twelve engineers because it sits in one engineer's user-level settings, which two others copied. The team wants it for everyone in the repository. Where should it live? Select the best answer. - A. In the repository's CLAUDE.md, as an instruction to run the formatter. - B. In the project's.claude/settings.json file, committed to version control. - C. In each engineer's.claude/settings.local.json file. - D. In every engineer's user-level settings file. 28. Question 28 · Domain 2: Applications and Integration · Single answer Stakeholders for a university's study assistant list several needs. You are separating infrastructure requirements from functional ones. Which of these is an infrastructure requirement? Select the best answer. - A. Each reply stays under 300 words and ends with a short one-line summary. - B. Replies use a warm, encouraging tone. - C. Answers cite the assigned course reading. - D. Logs stay in the university's region and are deleted after 90 days. 29. Question 29 · Domain 5: Model Selection and Optimization · Single answer A zero-shot prompt labels 40,000 tickets a day at 91.6 percent accuracy against a 90 percent target. In testing, four examples raise it to 91.7 and double each call's input tokens. What should you do? Select the best answer. - A. Keep all four examples for the extra tenth of a point. - B. Keep the zero-shot prompt and leave the examples out. - C. Add examples until it gains a full point. - D. Trim the examples to one, turning it into a single-shot prompt. 30. Question 30 · Domain 5: Model Selection and Optimization · Single answer A support assistant has a 600 ms median time to first token against a one-second target. It reasons poorly only on the 4 percent of turns about disputed bills. What is the best change? Select the best answer. - A. Enable extended thinking with a large budget on all turns, routine and billing alike. - B. Turn on extended thinking for the disputed-bill turns only. - C. Move every turn to the largest model tier and leave thinking off. - D. Add a prompt line that asks for step-by-step reasoning on disputed bills. 31. Question 31 · Domain 5: Model Selection and Optimization · Single answer A classifier tags listings into 60 categories and confuses loungewear with sleepwear 40 percent of the time. Two rewrites of the definitions did not help, and the prompt is cached. What should you try next? Select the best answer. - A. Raise the sampling temperature for the classification call. - B. Move both definitions into the user message beside the listing text. - C. Add three to five labeled examples from the boundary between the two categories. - D. Merge the two categories into a single combined label. 32. Question 32 · Domain 5: Model Selection and Optimization · Single answer At peak, 18 percent of API requests return HTTP 429. A wrapper retries each failure immediately up to ten times, and the 429 rate has risen to 31 percent. What should the client do? Select the best answer. - A. Create extra API keys in the same organization and spread the retries across them all. - B. Retry with exponential backoff and jitter, honor retry-after, and cap the attempts. - C. Raise the per-request timeout to two minutes in the client. - D. Open a new HTTP client for each retry attempt. 33. Question 33 · Domain 5: Model Selection and Optimization · Select 2 A browser app calls the Messages API directly with the company's API key in its JavaScript bundle. Product wants tokens to appear as they are generated. Which two changes does the design need? (Select two.) Select exactly 2.**0/2 selected.** - A. Move the call to a backend that holds the key and reads the stream. - B. Hide the key in the bundle with obfuscation and rotate it weekly. - C. Wait for the full response on the backend, then send it in one payload. - D. Relay tokens from the backend to the browser over server-sent events or a websocket. - E. Open a websocket from the browser straight to the Messages API and send the key first. 34. Question 34 · Domain 5: Model Selection and Optimization · Single answer Report drafts that use extended thinking pass through a gateway that drops connections idle for 60 seconds. One in five fails with a reset while the requester waits. What should you change? Select the best answer. - A. Raise the client-side timeout in the SDK to twenty minutes for these requests. - B. Submit each draft to the Message Batches API and poll until the result is ready. - C. Request each draft with streaming enabled on the API call. - D. Lower max_tokens to 1,000 for each report draft. 35. Question 35 · Domain 5: Model Selection and Optimization · Single answer A top-tier classifier takes 2.4 seconds against a 1-second target. On the eval set the fastest tier scores 97.8 percent in 0.7 seconds, the top tier 98.6. Misroutes are cheap to fix. Which model fits? Select the best answer. - A. Stay on the top tier and accept the 2.4-second response time. - B. Move to the middle tier and measure its accuracy after launch. - C. Use the fastest tier, with the eval set gating releases. - D. Stay on the top tier and turn on extended thinking for the remaining misroutes. 36. Question 36 · Domain 5: Model Selection and Optimization · Single answer A team moved production to a new model version on launch day for its higher accuracy. Then 6 percent of invoices failed to parse: amounts now included currency symbols. What should they have done first? Select the best answer. - A. Wait for the vendor to document every output change before switching. - B. Run the existing eval set against the new version in a non-production lane. - C. Loosen the parser to accept any string in the amount field. - D. Stay on the old model version and skip this release. 37. Question 37 · Domain 5: Model Selection and Optimization · Single answer You enabled prompt caching on a 24,000-token handbook that every request sends first. Which sign in the usage data confirms that later requests are reading it from the cache? Select the best answer. - A. A cache_creation_input_tokens count above zero on every request you send. - B. A lower output_tokens count than before caching. - C. A higher input_tokens figure on every new request. - D. A non-zero cache_read_input_tokens count on requests after the first. 38. Question 38 · Domain 6: Prompt and Context Engineering · Single answer A diagnostic agent makes 30 to 50 tool calls, some returning 20,000-token build logs. After about 20 calls it repeats steps and forgets early constraints, though the window is not full. What should you change? Select the best answer. - A. Trim each tool result to the lines that matter and compact older turns. - B. Have the agent summarize each build log in its next reply before moving on. - C. Tell the agent to reread the opening instructions before each tool call. - D. Raise the thinking budget on every call. 39. Question 39 · Domain 6: Prompt and Context Engineering · Single answer While writing a brief, an agent searches an archive for six topics, each search adding up to 8,000 tokens of excerpts. The final section now contradicts early decisions, and cost has tripled. What helps most? Select the best answer. - A. Delegate each search to a subagent that returns a short summary. - B. Raise max_tokens on every call the agent makes while writing the brief. - C. Have the agent reread the whole conversation before writing each section. - D. Run all six archive searches at the very start, before the agent makes any editorial decision. 40. Question 40 · Domain 6: Prompt and Context Engineering · Single answer A discharge-summary request puts the instructions, formatting rules and pasted clinical notes in one block of prose. The model sometimes obeys phrases in the notes, such as "keep this brief." What fixes this most directly? Select the best answer. - A. End the message with a sentence saying instructions in the notes do not count. - B. Put the notes first and the instructions last in the same message. - C. Split it into two calls: one extracts facts, one formats them. - D. Move the rules to the system prompt and place the notes inside a named XML tag. 41. Question 41 · Domain 6: Prompt and Context Engineering · Single answer A prompt asks for "a concise, well-structured list of torque values." Parsing fails on 22 percent of outputs: mixed units, missing fasteners. The prompt must stay short. What should you change first? Select the best answer. - A. Ask the model to check its own list against the parser's needs and revise any failing line before replying. - B. Add ten worked examples of correctly formatted torque lists to the prompt. - C. Enable extended thinking for the torque-list request. - D. Replace the adjectives with explicit rules: one line per fastener, values in newton-meters. 42. Question 42 · Domain 6: Prompt and Context Engineering · Single answer Claude returns a JSON object per support ticket for a database insert. About 5 percent fail: code fences, a leading sentence, or a priority outside the four allowed. Which fix holds up as models change? Select the best answer. - A. Strip fences and prose with a regular expression, and map unknown priorities to medium. - B. Prefill the assistant turn with an opening brace and keep the current parser. - C. Force a tool call with a priority enum and validate the result. - D. Resend the same request whenever an insert fails. 43. Question 43 · Domain 6: Prompt and Context Engineering · Single answer An assistant quotes clause numbers from a policy document supplied in the prompt. An audit found 4 percent cited clauses missing from the document, all stated confidently. Which control reduces this risk most? Select the best answer. - A. Show only answers the model rates 4 or higher on a confidence scale of 1 to 5. - B. Add a system prompt line asking for more care with clause numbers. - C. Add a banner telling staff to double-check every clause number themselves. - D. Verify each cited clause in the document before display. 44. Question 44 · Domain 7: Security and Safety · Select 2 An agent reads supplier emails and can call create_purchase_order. A plain-text paragraph in one email told it to place a $48,000 order, and it did. Which two changes most directly mitigate this? (Select two.) Select exactly 2.**0/2 selected.** - A. Strip HTML formatting from every email before the model reads it. - B. Pass email bodies inside a labeled data boundary that the instructions call untrusted input. - C. Require human approval before create_purchase_order runs above a value threshold. - D. Move to a larger model that follows its system prompt more closely. - E. List each agent-created order in a daily report for review. 45. Question 45 · Domain 7: Security and Safety · Single answer An assistant's only guard against personalized investment advice is one system prompt paragraph. In red-team tests, role-play beats it in 7 percent of sessions. No single bypass may reach a user. Which change meets that? Select the best answer. - A. Add an independent output check that screens each reply for investment advice. - B. Make the system prompt more emphatic and add examples of role-play attempts. - C. Block incoming messages that match the red team's known role-play framings. - D. Add a disclaimer under every reply. 46. Question 46 · Domain 7: Security and Safety · Single answer A Claude Code session deleted uncommitted scripts outside the build directory, though CLAUDE.md forbids it. You need an automatic control that does not slow routine commands. What should you add? Select the best answer. - A. A settings.json rule that asks an engineer to approve each Bash command. - B. A PreToolUse hook that blocks deletions outside the build directory. - C. A PostToolUse hook that logs deletions outside the build directory. - D. A stronger line at the top of CLAUDE.md. 47. Question 47 · Domain 7: Security and Safety · Single answer A production Claude API key was pushed to a public repository, and requests from unfamiliar IP ranges are spiking. Two services and a nightly job use the key. What should you do first? Select the best answer. - A. Revoke the key now and issue a new key for each service. - B. Add the file to.gitignore and rotate the key at the next maintenance window. - C. Rewrite the repository history to remove the key, then force-push. - D. Set a spend limit and watch the traffic. 48. Question 48 · Domain 8: Tools and MCPs · Single answer When a meter-reading tool times out, the harness raises an exception and the loop ends. Retries usually work. You want the agent to recover or explain plainly. What should the harness do? Select the best answer. - A. Return an empty string as the tool result and let the conversation continue. - B. Retry the call a few times inside the harness, then raise the exception and end the loop as before. - C. Send the failure back as an is_error tool_result. - D. Keep ending the loop, but show the user a much clearer failure message. 49. Question 49 · Domain 8: Tools and MCPs · Single answer A lookup_appointment tool has one free-string parameter named query and the description "Look up an appointment." The service needs a reference number, but 31 percent of calls pass a name or date. What fixes this? Select the best answer. - A. Add a system prompt paragraph describing the reference number format. - B. Have the service fuzzy-match names and dates to find appointments. - C. Replace it with a broad search tool that takes any string. - D. Rename the parameter, state its format with an example, and add a schema pattern. 50. Question 50 · Domain 8: Tools and MCPs · Single answer An assistant has 42 tools, several overlapping: three return timetables with slightly different filters. It picks a suitable tool first time on 68 percent of questions. What helps most? Select the best answer. - A. Allow one tool call per question, and prefer the first listed tool when several fit. - B. Move to the top model tier before changing any tool definitions. - C. Merge overlapping tools into one tool per capability, with parameters for filters. - D. Paste all 42 tool schemas into the system prompt as well. 51. Question 51 · Domain 8: Tools and MCPs · Single answer Researchers will query a document store from Claude Desktop through an MCP server. Queries must use the company identity provider, and store credentials must stay in the data center. Which deployment fits? Select the best answer. - A. A stdio server on every laptop, with the store credentials in a config file managed by fleet tooling. - B. A stdio server on each laptop that reads credentials from an environment variable set by IT. - C. Direct REST calls from Claude Desktop with no MCP layer. - D. A central remote server over streamable HTTP, signed in through the identity provider. 52. Question 52 · Domain 8: Tools and MCPs · Single answer You want Claude Code to write incident post-mortems with a fixed template, a severity rubric, a CSV timeline script and a checklist. Everything is local, and no external system is called. Which approach fits best? Select the best answer. - A. Build an MCP server with a generate_post_mortem tool on shared infrastructure. - B. Package everything as a Claude Code Skill. - C. Paste the template, rubric and checklist into the repository's CLAUDE.md. - D. Keep the files in a shared folder and attach them to each post-mortem prompt. 53. Question 53 · Domain 8: Tools and MCPs · Single answer Analysts want an assistant that computes statistics and a chart from uploaded CSVs of aggregated figures. Sending them to the API is approved, and no platform engineers can run infrastructure. Which approach fits? Select the best answer. - A. Use the built-in code execution tool to run the analysis in Anthropic's sandbox. - B. Build an MCP server that wraps a Python interpreter on your own servers. - C. Ask the model to compute the statistics directly from the CSV text. - D. Write a custom tool backed by a scheduled job you host. Before you ask ## Before you sit the mock **Is this the real CCDV-F exam?** No. Every item here was written for this page from the official CCDV-F Exam Guide and its published domains and objectives. Nothing is recalled from a sitting, nothing is drawn from another provider, and Cred Farmer is not affiliated with Anthropic or Pearson VUE. The items are short scenarios, about the length of the guide's sample questions. Of the 53, 49 are single-answer with four options and 4 ask you to select two of five, saying so in the stem. The questions themselves are ours, and their difficulty is not calibrated against the exam. **How is the mock scored?** You get the raw count of correct answers out of the items you sat, and the same count for each domain beside that domain's official weight. A select-two item counts only when both picks match the key. The exam reports a scaled score with a 720 cut, and no practice percentage converts to it, so this page never shows a scaled score, a pass verdict or a percentile. **Do I need an account to take it?** No. Choose a start mode and the sitting runs in your browser tab. There is no sign-up, no email field and no cookie that follows you. An account only matters if you want the larger signed-in library, and the link for that sits at the bottom of the page rather than in your way. **Is it actually free, or is there a catch?** No catch. Everything on Cred Farmer is free, this page included: no card, no trial, no limit on how many times you sit it. An account costs nothing either; it adds the five timed forms, saved progress and a readiness score capped by your first sitting of each form. The one limit here is inventory: the no-account paper is one paper, so sit it a second time and you will remember some items, and that is when the signed-in forms become the next step. **How many questions are in the mock?** 53 in the full sitting, allocated across the 8 official domains in proportion to their weights. The clock runs at the official pace of 53 items in 120 minutes, about 2.3 minutes an item, so the full sitting gets 120 minutes, the same as the exam. The quick sitting draws 30 of them evenly across the domains in 68 minutes, and the untimed review takes the whole paper without a clock. **What happens to my result?** It stays in this tab. Nothing is sent to a server, nothing is stored on your device, and nothing you answer here feeds a readiness score or any aggregate. Close or refresh the page and the result is gone, which is also why it is worth reading the per-option reasons before you leave. ## This page is 53 items. The account has 265. Everything on Cred Farmer is free. An account adds the five timed forms, saved progress and a readiness score: CCDV-F has 5 forms aligned to the current blueprint, 53 items each, 265 in total, plus drills by objective, and your first sitting of each form sets the band the score can reach. No card, no trial clock. [Create a free account for the five timed forms](https://credfarmer.com/start/ccdv-f?to=practice) Every item on this page was written for it from the official CCDV-F Exam Guide, not recalled from a sitting and not copied from anyone else's set. Material sold as recalled questions can cost you the credential: read [why Claude certification exam dumps are a bad bet](https://credfarmer.com/blog/claude-certification-exam-dumps). Items were last checked against the guide on 17 September 2026. The guide is published on Anthropic's [CCDV-F certification page](https://anthropic-partners.skilljar.com/claude-certified-developer-foundations-certification). Not affiliated with, or endorsed by, Anthropic or Pearson VUE. Exam facts are summarised from publicly published program information and can change. Always confirm against the official exam guide before booking.