CCA-F Practice Questions — Page 9

Question 81

Scenario: Tool Design and MCP Integration

An MCP tool fails because the orders API timed out. It currently returns `{"isError": true, "content": "Operation failed"}`. The coordinator cannot decide whether to retry, change the query, or escalate.

What should the error response include?

A.Just a longer human-readable message describing the failure.
B.Structured metadata: `errorCategory` (e.g., transient), `isRetryable`, a clear message, the attempted query, and any partial results.
C.Nothing extra; the agent should retry every `isError` response automatically.
D.An empty success response so the workflow keeps moving.

Question 82

Open question ↗

Scenario: Tool Design and MCP Integration

A refund tool rejects a request because the order is outside the 30-day return windowâ€”a policy rule, not a temporary glitch.

How should this error be represented?

A.`errorCategory: "business"` with `isRetryable: false` and a clear user-facing explanation.
B.`errorCategory: "transient"` with `isRetryable: true` so the agent retries with backoff.
C.A generic `"Operation failed"` message.
D.Return success with an empty result to avoid confusing the agent.

Question 83

Open question ↗

Scenario: Tool Design and MCP Integration

A team wants a shared GitHub MCP server available to everyone who clones the repo, with the token kept out of version control.

How should this be configured?

A.Add the server to project `.mcp.json` and reference the token via an environment variable like `${GITHUB_TOKEN}`.
B.Add the server to `~/.claude.json` with the token hard-coded.
C.Paste the raw token directly into `.mcp.json` and commit it.
D.Put the server config in the root CLAUDE.md so the team sees it.

Question 84

Open question ↗

Scenario: Tool Design and MCP Integration

An agent must understand the structure of a large project-management dataset (all tasks, hierarchy, statuses) to answer questions. Today it makes many exploratory tool calls just to learn what data exists, wasting context.

Which MCP capability removes the exploratory calls?

A.Expose the catalog/schema as an MCP resource the agent can read for an immediate map of available data.
B.Add more tools so each exploratory question has a dedicated tool.
C.Increase `max_tokens` so the exploratory calls fit in context.
D.Force `tool_choice: "any"` to make exploration faster.

Question 85

Open question ↗

Scenario: Tool Design and MCP Integration

An agent has both a built-in `Read` tool and an MCP tool `load_internal_doc` that fetches authoritative, access-controlled internal documents. The agent keeps using `Read` and misses the internal source.

What is the best fix?

A.Strengthen the MCP tool's description to highlight the unique, authoritative data it provides that built-in tools cannot access.
B.Remove the `Read` tool entirely from every agent in the system.
C.Rename `load_internal_doc` to `Read2` so the model treats it like the built-in.
D.Set temperature to 0 so the model deterministically prefers MCP tools.

Question 86

Open question ↗

Scenario: Tool Design and MCP Integration

In a Claude Code session you need to find every file matching `**/*.test.ts`, then locate the definition of a function named `calculateTax` inside the codebase.

Which built-in tools fit these two tasks?

A.Use `Glob` to find files by the name pattern, and `Grep` to search file contents for the function.
B.Use `Grep` to find files by pattern, and `Glob` to search contents.
C.Use `Read` on every file to do both.
D.Use `Bash` with `find` and `cat` since no dedicated tools exist.

Question 87

Open question ↗

Scenario: Tool Design and MCP Integration

A general `fetch_url` tool is used by a document-analysis subagent, but the agent sometimes fetches unrelated web pages and pollutes its context. You want to constrain it to loading vetted documents only.

What is the recommended approach?

A.Replace the broad `fetch_url` with a constrained `load_document` tool scoped to approved document sources.
B.Keep `fetch_url` but add a system-prompt rule telling the agent not to browse the open web.
C.Give the agent every available tool so it has options.
D.Set `tool_choice: "auto"` and rely on the model's judgment.

Question 88

Open question ↗

Scenario: Claude Code Configuration and Workflows -

A new team member clones the repo but Claude Code doesn't apply the project's coding standards. Investigation shows the standards were written in the senior developer's `~/.claude/CLAUDE.md`.

What is the fix?

A.Move the standards into the project's `.claude/CLAUDE.md` (or root `CLAUDE.md`) so they are shared via version control.
B.Ask every team member to copy the senior developer's `~/.claude/CLAUDE.md`.
C.Put the standards in `.claude/config.json`.
D.Leave it; user-level CLAUDE.md applies to the whole team automatically.

Question 89

Open question ↗

Scenario: Claude Code Configuration and Workflows -

A monorepo has many packages. The root CLAUDE.md has grown huge because it duplicates testing and coding standards inline. You want each package's CLAUDE.md to pull in only the relevant shared standards without duplication.

Which mechanism should you use?

A.The `@path` import syntax (e.g., `@./standards/testing.md`) to reference shared files modularly.
B.Copy-paste the standards into each package's CLAUDE.md.
C.Put all standards in a single skill and invoke it manually per package.
D.Store standards in `.claude/config.json` and reference by key.

Question 90

Open question ↗

Scenario: Claude Code Configuration and Workflows -

Test files are scattered across many directories. You want a testing convention to load automatically only when Claude Code edits a test file, without bloating context for unrelated work.

What is the best approach?

A.A `.claude/rules/` file with YAML frontmatter `paths: ["**/*.test.ts", "**/*.test.tsx"]`.
B.A directory-level CLAUDE.md placed in one tests folder.
C.All testing rules in the root CLAUDE.md so they always load.
D.A skill that the developer must remember to run before editing tests.