Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 39

Question 380

An MCP client connects to a server implementing a different protocol revision. What should happen before normal feature use?

  1. Treat an open TCP connection as proof of feature compatibility.
  2. Assume all protocol revisions are interchangeable.
  3. Perform initialization and negotiate compatible protocol and capabilities.
  4. Call every known tool to discover compatibility through side effects.
  5. Silently reinterpret unknown message fields as tool arguments.

Explanation

Correct Answer

C:

Agree on the protocol revision and supported capabilities before relying on particular messages; an open connection alone proves only transport reachability.

Incorrect Answers

A:

Network connectivity alone does not establish agreement on the application protocol.

B:

Version differences can affect supported messages or behavior and must be handled explicitly.

D:

Invoking tools is not a substitute for protocol negotiation and can execute unintended operations.

E:

Arbitrary reinterpretation can corrupt semantics and bypass the negotiated interface.

Question 381

An MCP server exposes several tools. How should a client learn the tool names, descriptions, and input schemas?

  1. Read only the transport port number.
  2. Infer the full tool list from the server's hostname.
  3. Generate plausible names and execute them until one succeeds.
  4. Use the protocol's tool-listing mechanism and honor the returned contracts.
  5. Assume every MCP server implements a universal set of database tools.

Explanation

Correct Answer

D:

Discovery supplies the actual supported tool definitions, which the client can present and validate appropriately.

Incorrect Answers

A:

The port identifies a network endpoint, not the available application-level tools.

B:

A hostname does not encode the server's capabilities or input schemas.

C:

Guessing bypasses reliable discovery and can trigger errors or unintended operations.

E:

MCP standardizes interaction, while each server chooses the capabilities it exposes.

Question 382

A local MCP server is launched as a child process using stdio. Another deployment must support remote clients over the network. Which change is appropriate?

  1. Place the server's private filesystem path in the model prompt.
  2. Disable protocol initialization for remote clients.
  3. Keep stdio unchanged and assume remote clients can open the local pipes.
  4. Expose the child process's stdout as an unauthenticated public webpage.
  5. Use a supported remote transport and add the required authentication and deployment controls.

Explanation

Correct Answer

E:

Process pipes are local communication channels; remote access needs a network transport and suitable security configuration.

Incorrect Answers

A:

A path description does not establish a network connection or authenticated protocol session.

B:

Remote deployment still requires protocol agreement and supported capability handling.

C:

Local process pipes do not automatically become accessible to remote clients.

D:

Publishing output is neither a complete MCP transport nor an adequate remote security design.

Question 383

An MCP server accepts authenticated client requests and calls a downstream business API. What must it enforce?

  1. Use one unrestricted administrator credential for all callers.
  2. Validate access for the MCP service and use appropriately authorized downstream credentials.
  3. Forward every incoming bearer token to any downstream service.
  4. Trust a username supplied in free-form tool arguments.
  5. Skip authorization because MCP standardizes message structure.

Explanation

Correct Answer

B:

The server must preserve the intended authorization boundaries rather than treating any received token as universal authority.

Incorrect Answers

A:

A universal privileged identity erases caller boundaries and increases the impact of unauthorized requests.

C:

Tokens are issued for particular audiences and scopes; blind passthrough breaks that boundary and can expose credentials.

D:

A claimed username is not authenticated identity or proof of permission.

E:

Protocol interoperability does not grant users permission to access underlying data or actions.

Question 384

An MCP resource changes after a client reads it. The client wants automatic change notifications. What must it verify before relying on them?

  1. That reading a URI once always subscribes to all future changes.
  2. That every notification is a new authorization grant.
  3. That cached content remains current even when no subscription exists.
  4. That the relevant subscription and notification capabilities are supported and correctly used.

Explanation

Correct Answer

D:

A shared protocol does not make every optional feature universal; the client must establish the server's advertised support and subscribe as required.

Incorrect Answers

A:

A one-time read does not by itself establish an active subscription or universal notification guarantee.

B:

Change events convey state information, not permission to access additional resources.

C:

Without refresh or change evidence, previously read content can become stale.

Question 385

An MCP client sends notifications/initialized without a JSON-RPC id and then waits indefinitely for a matching response. What should the client change?

  1. Use the notification as the only initialization message and skip version negotiation.
  2. Retry the notification until the server invents a response identifier.
  3. Assume the server is disconnected whenever it sends no notification response.
  4. Treat it as a notification that does not require a response.

Explanation

Correct Answer

D:

The initialization request has its own response, but the subsequent initialized message is a notification. Waiting for a response to an id-less notification introduces a wait the protocol does not fulfill.

Incorrect Answers

A:

The notification follows successful initialization; it does not replace the request-response exchange that establishes protocol agreement.

B:

Repeated notifications do not turn the message into a request. The client must follow the message type's response semantics.

C:

Silence is expected for this notification and is not, by itself, evidence of a transport failure.

Question 386

A Claude Code Skill performs a release procedure that should run only when a user deliberately invokes it. Which documented setting fits that intent?

  1. Remove the Skill's procedure but keep its name.
  2. Set user-invocable: false and assume that blocks automatic model use.
  3. Set disable-model-invocation: true in the Skill's frontmatter.
  4. Put an API key in the Skill name.
  5. Put the phrase "release tool" in every prompt.

Explanation

Correct Answer

C:

This prevents automatic model invocation while leaving the Skill available for explicit user invocation through its supported interface.

Incorrect Answers

A:

Removing the workflow prevents useful execution instead of controlling who can trigger it.

B:

Hiding a Skill from user invocation addresses a different control; it does not express the desired user-only trigger behavior.

D:

Credentials do not configure invocation policy and should not be exposed as names.

E:

Repeated wording does not configure the Skill's invocation policy and may encourage unnecessary activation.

Question 387

A host already provides an approved file-reading tool that satisfies a project requirement. What is the best initial implementation choice?

  1. Grant unrestricted shell access solely because a tool already exists.
  2. Build a second identical file-reading service before testing the existing one.
  3. Ask the model to infer file contents from filenames.
  4. Use the existing capability and verify it meets the required contract.
  5. Copy every possible file into the system prompt permanently.

Explanation

Correct Answer

D:

Reusing a suitable approved tool avoids unnecessary integration and maintenance while still requiring fit validation.

Incorrect Answers

A:

Broader execution permissions are not required to use a sufficient file-reading capability.

B:

Duplicate infrastructure adds complexity without evidence of a capability gap.

C:

Inference cannot replace reading the actual source when the task depends on its content.

E:

A permanent bulk copy becomes stale and wastes context compared with targeted access.

Question 388

An application's backend exposes a proprietary pricing calculation unavailable through existing approved tools. What is the most direct integration?

  1. Publish the entire backend source and credentials in the prompt.
  2. Wrap the calculation in a narrowly defined tool with validated inputs and documented outputs.
  3. Change the model name until the missing API appears.
  4. Use a static cached answer for every pricing request.
  5. Ask Claude to recreate the secret formula from examples at runtime.

Explanation

Correct Answer

B:

A focused contract gives the model access to the missing capability while retaining backend control over execution.

Incorrect Answers

A:

Broad disclosure is unnecessary for calling one calculation and exposes unrelated implementation and secrets.

C:

Model selection does not create access to an application-specific backend function.

D:

A constant result cannot satisfy input-dependent pricing calculations.

E:

Approximation does not guarantee the proprietary service's authoritative pricing behavior.

Question 389

A Claude Code SKILL.md starts with a Markdown heading, followed later by a YAML-looking block containing description. The author expects that block to configure Skill metadata. What should be corrected?

  1. Keep the block below the heading but add Markdown code fences around it.
  2. Put the YAML frontmatter at the beginning of the file.
  3. Put the same YAML only in a linked reference file.
  4. Rename the body section Metadata without moving the YAML block.
  5. Repeat the description in several ordinary body paragraphs.

Explanation

Correct Answer

B:

Claude Code recognizes Skill frontmatter when the opening delimiter is on the first line. A later block can be treated as body text rather than the metadata the author intended to configure.

Incorrect Answers

A:

A fenced YAML example in the body is still document content. The metadata parser expects frontmatter at the start of the file.

C:

Supporting references do not replace the Skill entry file's frontmatter. Metadata must be declared in the expected location in SKILL.md.

D:

A section label does not make ordinary body content into frontmatter. Position and frontmatter syntax determine metadata parsing.

E:

Repeating text does not turn a later YAML-looking block into recognized frontmatter. The file structure must satisfy the metadata parser's boundary rule.

Question 390

A Skill needs a short procedure plus a long reference manual used only for rare cases. How should it organize this content?

  1. Keep core instructions concise and reference supporting material for use when needed.
  2. Replace all examples with links that have no applicability explanation.
  3. Paste the entire manual into every user message.
  4. Omit the reference entirely because it is long.
  5. Store the manual as a secret environment variable.

Explanation

Correct Answer

A:

Separating routine guidance from optional detail preserves context while keeping specialized information available.

Incorrect Answers

B:

References still need enough context for the assistant to know when and why to consult them.

C:

Repeated bulk context increases cost and distracts from the current task.

D:

Removing needed specialized guidance creates avoidable gaps for the rare cases that depend on it.

E:

A reference document is not a credential, and environment variables are not a useful retrieval interface for long guidance.

Claude Certified Developer - Foundations (CCDV-F) Practice Tests