Question 120
A team connects Claude Code to an internal MCP document service for collaborative research. Which THREE practices support a reliable and secure integration? Choose THREE.
- Paste confidential service credentials into the model's visible prompt
- Preserve structured tool errors so Claude can recognize failed retrievals
- Train a new model every time the MCP resource catalog changes
- Configure an appropriate client transport and reachable MCP endpoint
- Limit exposed capabilities and enforce user permissions on the service
- Assume a connected server makes every private resource public by default
Explanation
Correct Answers
B:Transport success does not imply business success. Actionable error messages support retries, clarification and honest uncertainty.
D:The client must establish an actual server connection before advertised tools or resources can be discovered. Prompt text alone is not connectivity.
E:MCP standardization does not replace authorization. The server and client should prevent unrelated destructive actions for a read-only role.
Incorrect Answers
A:Secrets should be managed by the runtime and authentication layer. Exposing them as prompt context creates unnecessary disclosure risk.
C:MCP advertises current capabilities and data dynamically. Model retraining is not the mechanism for refreshing client discovery.
F:Connection and entitlement are different properties. The backend still controls what a user may access.
Question 121
A connected MCP tool changes its input schema, and calls begin failing. What should the client team check first?
- Retry the same malformed parameters indefinitely
- Assume the language model's weights must be retrained
- Suppress all validation failures so calls appear successful
- Refresh discovered tools and reconcile the current schema
Explanation
Correct Answer
D:MCP clients should use the server's current capability definitions. A stale argument contract can explain systematic validation errors.
Incorrect Answers
A:Repeated calls cannot make an obsolete schema valid. The interface contract needs updating.
B:Tool contracts are integration metadata. Schema changes are resolved in discovery/configuration and client usage, not model training.
C:Ignoring incompatible inputs hides actual execution errors. It does not restore the tool's behavior.
Question 122
A company wants an MCP server to expose static policy documents for reading, not only executable actions. Which MCP concept is useful?
- Assume resource names automatically make documents public
- Put all private policies verbatim into global instructions for every project
- Publish resources that clients can retrieve according to their permissions
- Convert every document into a destructive update tool
Explanation
Correct Answer
C:MCP resources provide a standardized way to expose contextual data. They complement tool operations and should preserve access boundaries.
Incorrect Answers
A:Resource exposure remains governed by server implementation and authorization. Naming is not access control.
B:Universal context inclusion raises confidentiality and maintenance issues. Scoped retrieval is more appropriate.
D:Mutation interfaces are unnecessary for read-only policy text. They introduce side effects and complicate discovery.
Question 123
A production tool client shows an MCP server as connected but no relevant tools are available. What should troubleshooting focus on?
- Ask the model to invent tool names and call them
- Increase Claude's response length because tools are missing
- Inspect capability discovery and client-side tool scopes
- Delete all other project instructions before checking the connection
Explanation
Correct Answer
C:Network connectivity alone does not guarantee the expected tools are registered or permitted. The visible capability catalog must be checked.
Incorrect Answers
A:Calls must match discovered interfaces. Fabricated tool identifiers cannot execute successfully.
B:Token budget does not make unpublished server capabilities appear. The integration layer is the likely fault.
D:Instruction clutter may influence selection, but absence of discovered tools is a configuration problem first.
Question 124
An MCP server returns a tool exception, but the client describes the operation as successful. What interoperability aspect needs correction?
- Preserve the server's tool error status and explanatory content end to end
- Treat every HTTP 200 response as proof that a tool operation succeeded
- Hide the server result to save context tokens
- Retry all operations immediately without checking side effects
Explanation
Correct Answer
A:The client must transmit execution failure to the model rather than treating transport success as business success. Accurate status enables recovery.
Incorrect Answers
B:MCP tool-level errors may be represented inside a valid protocol response. Transport status and operation status are distinct.
C:Removing the observation prevents the agent from recognizing failure. Concise structured errors are preferable.
D:Some tools mutate data, and blind retries can duplicate actions. A truthful error category and idempotency check are needed.
Question 125
A code analysis agent uses an MCP endpoint that also advertises billing modifications. Which client integration choice best limits unintended actions?
- Expose or permit only the code-analysis capabilities for that agent
- Require a billing tool call to begin every code inspection
- Mirror the billing data into code prompts to avoid using the endpoint
- Let the agent see everything and append a gentle warning
Explanation
Correct Answer
A:Capability filtering should match the worker's legitimate role. Billing mutations are unrelated to source inspection and should remain unavailable.
Incorrect Answers
B:Forcing an irrelevant mutation increases risk and does not help analysis. The chosen tools must match task needs.
C:Copying sensitive data into context expands exposure without solving access control. The source task needs no billing information.
D:Language instructions cannot enforce least privilege by themselves. Tool exposure or execution authorization should restrict the role.
Question 126
A code assistant must find all references to a deprecated function without changing files. Which tool behavior fits the task?
- Use targeted repository search or read-only grep
- Use Edit to replace every occurrence immediately
- Ask the model to list references from general training memory
- Run a project-wide source formatter before searching usages
- Use a shell command that deletes files matching the function name
Explanation
Correct Answer
A:Search tools are designed to enumerate occurrences efficiently. Read-only inspection avoids unintended edits while building an impact list.
Incorrect Answers
B:Mutation before inspection may change unrelated uses and violates the read-only requirement. First identify the exact references.
C:The current repository may differ from any learned examples. Actual code search is necessary.
D:Formatting changes files and can create unrelated diffs. A reference inventory should be collected with read-only operations.
E:Deletion is unrelated to symbol search and is destructive. It cannot produce a safe usage inventory.
Question 127
An assistant has identified a small change in a single tracked source file. Which built-in capability is the most direct fit?
- Replace the complete repository from a generated description
- Use only read or grep operations and claim the change is applied
- Use the Edit tool with a targeted change and verify the diff
- Ask a web-search tool to rewrite the local repository
Explanation
Correct Answer
C:A focused edit preserves surrounding content and is easier to review than regenerating the entire file. Validation should follow mutation.
Incorrect Answers
A:Broad regeneration is unnecessary and risky for a small localized change. It can remove unrelated working code.
B:Inspection does not modify source files. The requested update needs a controlled write operation.
D:External web retrieval cannot directly perform a local file edit. It addresses a different environment.
Question 128
A developer needs to run the project's existing unit tests after an edit. Which tool category is required?
- Use only a file-reading tool to infer that tests passed
- Use a search tool to count test filenames as proof of success
- Run the test in a controlled shell and check its exit code
- Ask Claude to produce a reassuring success sentence
Explanation
Correct Answer
C:Tests are executable workflows. Capturing output and failure codes supplies concrete evidence about the change.
Incorrect Answers
A:Reading source cannot establish runtime test results. Execution is required for the stated acceptance check.
B:Test existence does not indicate whether tests currently pass. A test runner must evaluate them.
D:Generated assurance is not execution evidence. The acceptance criterion is the actual test result.
Question 129
An analyst must understand the shape of a large unfamiliar repository before editing. Which approach is safest?
- Disable source inspection tools to reduce context usage
- Use directory listing, targeted search and file reads to map relevant areas
- Read only the largest file and assume it describes the whole system
- Apply a repo-wide rename before locating dependencies
Explanation
Correct Answer
B:Broad orientation followed by focused inspection reveals architecture with minimal side effects. Editing can begin after ownership is understood.
Incorrect Answers
A:Less evidence can make edits less reliable. Targeted inspection controls cost without removing visibility.
C:File size is not an architecture map. Relevant code may live in smaller modules and configuration files.
D:Mutation without understanding references can break unrelated components. Discovery should precede a broad refactor.
Question 130
A task is to update one exact configuration value, and the file contents are already known. What tool choice minimizes unwanted changes?
- Use web fetch to inspect an unrelated public template
- Report completion after printing the proposed line to the console
- Apply a precise file edit and check the resulting diff
- Regenerate the entire configuration from a high-level guess
Explanation
Correct Answer
C:A targeted replacement limits the mutation footprint. Verification ensures the correct occurrence changed and no unrelated content moved.
Incorrect Answers
A:An external template may not match local constraints. The requested change concerns a known repository file.
B:Printing a suggestion is not a file modification. The on-disk artifact must be updated and checked.
D:Whole-file generation risks losing comments and settings. The existing file already provides a reliable baseline.