Anthropic CCA-F (Claude Certified Architect - Foundations (CCAR-F)) Practice Test 21

Question 200

A team changes an MCP connection's transport. What must remain distinguished from that choice?

  1. Whether connection failures can occur.
  2. Whether the client can reach the server.
  3. How the client detects a disconnected message channel.
  4. The framing and delivery mechanisms used to exchange protocol messages over the selected connection.
  5. The tool's business semantics and authorization requirements.

Explanation

Correct Answer

E:

A transport change alters message carriage and connection behavior. The operation's effects and access rules remain application contracts that must be evaluated separately.

Incorrect Answers

A:

Availability concerns are relevant to transport.

B:

Reachability is also a transport concern.

C:

Connection lifecycle and failure detection belong to transport handling. They do not define whether a tool's underlying business operation is permitted.

D:

Message carriage is directly part of the transport choice. The tool's business effects and access rules remain separate responsibilities.

Question 201

A client and server support different protocol versions. What is safer than silently assuming compatibility?

  1. Ignore every server error.
  2. Use the newest field names unconditionally while keeping the old negotiated protocol behavior.
  3. Send arbitrary messages until something appears to work.
  4. Use the protocol's supported version/capability handling and fail clearly when incompatible.

Explanation

Correct Answer

D:

Explicit compatibility checks prevent the client from relying on operations or fields the server does not support. Version and capability handling establishes the features both peers can actually use. A mismatch should remain visible rather than being hidden behind messages that one side cannot interpret.

Incorrect Answers

A:

Suppressed incompatibilities can become misleading success paths.

B:

Mixing versions without compatibility handling can produce messages the peer cannot interpret.

C:

Trial-and-error can conceal unsupported semantics and incomplete behavior.

Question 202

In MCP version 2025-11-25, a known tool executes a valid request but cannot complete a business operation. How should this normally differ from an unknown protocol method?

  1. Report tool-execution failure separately from protocol-method errors.
  2. Return an empty success object for every failed execution.
  3. Close the connection without any result whenever the operation is unsuccessful.
  4. Treat every business failure as an unknown-method protocol error.

Explanation

Correct Answer

A:

Separating layers lets clients distinguish a tool's business failure from an invalid protocol interaction. A recognized method can fail during its business operation without making the protocol request itself invalid. Preserving that distinction gives the caller a meaningful recovery category.

Incorrect Answers

B:

A false success hides information the caller needs for recovery.

C:

Dropping the connection discards useful failure semantics for an otherwise valid request.

D:

The method was recognized; the failure occurred inside the requested tool operation.

Question 203

A tool parameter named query is used for both a search expression and an exact object ID in different modes. What improves usability?

  1. Separate or clearly discriminate the modes and document each parameter's meaning.
  2. Let the backend silently reinterpret any string.
  3. Use the same vague description for every mode.
  4. Keep one unconstrained string parameter and make downstream services infer the mode.

Explanation

Correct Answer

A:

Overloaded fields encourage wrong assumptions. Explicit mode contracts make valid inputs and expected behavior easier to distinguish. Mode-specific fields or a discriminator remove uncertainty about how the same string will be interpreted. This prevents exact identifiers from being accidentally treated as broad search expressions.

Incorrect Answers

B:

Unpredictable coercion makes errors harder to detect and reason about.

C:

That preserves the confusion.

D:

Shifting ambiguity into execution leaves the caller unable to express a clear contract for search versus exact lookup.

Question 204

A tool's HTTP request returns 200, but its business payload says the operation was rejected. What should the agent report?

  1. Success based solely on HTTP 200, ignoring the operation status in the body.
  2. The operation's actual rejected status and relevant reason.
  3. Retry the operation as a network timeout without reading the application's rejection reason.
  4. Tell the user only the numeric HTTP status and omit the rejected business outcome.

Explanation

Correct Answer

B:

HTTP 200 indicates that the HTTP request was handled successfully under that endpoint's convention. The payload explicitly reports rejection of the requested business operation, so that is the outcome the agent must communicate.

Incorrect Answers

A:

Transport-level success can carry an application-level rejection that must govern the reported business outcome.

C:

The request received a completed response. Recovery should follow the stated business rejection rather than misclassifying it as an unobserved transport outcome.

D:

The numeric status alone would conceal the result the user needs. The application explicitly reported that the requested operation was not accepted.

Question 205

A tool deprecates a field but still accepts it temporarily. What should maintained callers do?

  1. Choose the replacement field by matching its name to similar fields in unrelated tools.
  2. Continue indefinitely without checking migration guidance.
  3. Submit both old and new fields with conflicting values and let each backend choose.
  4. Follow the documented replacement and verify behavior during migration.

Explanation

Correct Answer

D:

A controlled transition preserves semantics while removing reliance on a retiring contract. A deprecation period supports transition rather than indefinite reliance. Testing the documented replacement verifies that the migrated caller preserves the intended behavior.

Incorrect Answers

A:

Similar naming is not evidence that this API supports the same parameter or interprets it equivalently.

B:

Temporary compatibility may end and can conceal future breakage.

C:

Conflicting representations create ambiguous behavior during migration rather than a verified transition.

Question 206

A tool parameter force can bypass checks, but its description is absent. What should happen before allowing the model to use it?

  1. Remove all backend validation whenever force appears.
  2. Set it true by default for convenience.
  3. Document its exact effects, constrain its use, and enforce prerequisites.
  4. Assume force only improves performance because it is an optional boolean.

Explanation

Correct Answer

C:

Risk-bearing switches need explicit semantics and authorization. The model should not infer what checks are bypassed. A bypass parameter can materially change safeguards or overwrite behavior. Its type alone does not explain when it is permitted, so both documentation and implementation checks are needed.

Incorrect Answers

A:

The flag cannot justify abandoning unrelated safety or access rules.

B:

An unexplained bypass can expand consequences unexpectedly.

D:

Optionality and type do not establish the flag's effects or justify bypassing checks.

Question 207

An MCP server receives a bearer token intended for a different resource server. Under the 2025-11-25 authorization model, what should it do?

  1. Validate that the token is intended for this resource and reject an audience mismatch.
  2. Forward it to every downstream API until one accepts it.
  3. Accept it if the token string looks long enough.
  4. Trust any token accompanied by a valid tool name.

Explanation

Correct Answer

A:

Audience validation prevents a token issued for another service from being misused at this one. A credential issued for another audience should not be accepted just because it is otherwise well formed. Separate downstream APIs require credentials valid for their own resource boundaries.

Incorrect Answers

B:

Unrestricted token forwarding crosses audience boundaries and exposes credentials unnecessarily.

C:

Token length does not establish its intended audience or validity for this server.

D:

Tool selection does not make a credential valid for the receiving resource.

Question 208

A developer wants Claude Code to start with her preferred model in every repository on her laptop. No team policy sets the model. Where should she persist that preference?

  1. Her user-level Claude Code settings.
  2. The current session's prompt only.
  3. A managed settings file distributed to the organization.
  4. The README of each repository.
  5. Commit the preference into every repository's shared settings so colleagues receive the same default whenever they clone one of her projects.

Explanation

Correct Answer

A:

User settings apply to her projects on that machine and fit a personal default. A repository's shared settings would affect its team rather than all of her personal projects. Higher-precedence policies can still constrain the effective value of this preference.

Incorrect Answers

B:

A session instruction does not persist the configuration across future repository sessions.

C:

Organization policy would impose a personal preference on other developers unnecessarily.

D:

A README documents the project but does not serve as the user's cross-project settings file.

E:

The requirement is personal and cross-project. Repeated shared-project edits impose the preference on others and require maintenance in every repository.

Question 209

A team has reviewed a Claude Code hook configuration and wants every clone of its repository to receive it. What should the maintainer do?

  1. Place the configuration only in her home-directory settings.
  2. Commit the hook configuration in the shared project settings.
  3. Put it in the project's local settings and leave it untracked.
  4. Add the hook's name to a commit message.

Explanation

Correct Answer

B:

Versioned project settings distribute the team's configuration and make changes reviewable. Keeping the configuration in version control also lets reviewers inspect future hook changes. A maintainer's home-directory setting would remain local to that maintainer.

Incorrect Answers

A:

Her local user settings do not travel with the team's repository clones.

C:

An untracked personal file will not distribute the reviewed configuration to colleagues.

D:

A commit message describes changes but does not register an executable hook.

Question 210

A developer's user setting allows an action that an applicable managed policy denies. The action is still blocked. What should the developer investigate first?

  1. Whether a contradictory project instruction says the managed restriction may be ignored.
  2. Whether the same allow rule appears twice in the user file.
  3. Whether the conversation has enough remaining tokens.
  4. The effective managed policy and its precedence.

Explanation

Correct Answer

D:

Organization-enforced restrictions can override personal settings, so the effective policy explains the block. Troubleshooting should examine the effective sources rather than repeatedly adding a lower-precedence allow rule. A blocked action can be the expected result of the organization's policy.

Incorrect Answers

A:

Project prose does not override an enforced organization-level configuration policy.

B:

Duplicating a personal rule does not give it authority over managed restrictions.

C:

Context capacity does not determine which configuration policy takes precedence.

Claude Certified Architect - Foundations (CCAR-F) (CCA-F) Practice Tests