CCA-F start here · Domain 2 · MCP mind maps

MCP mind maps

Everything the CCA-F exam asks about MCP, on one page. Most of it comes down to four questions. Each question gets a mind map in the form "what you see, where the fault is, the fix", plus the rule to remember, the exam-guide line it comes from, and one practice question.

Domain 2 · 18%Guide 2.1–2.4About 20 minutes

What the exam does not test: building, deploying or hosting MCP servers, OAuth, and transports. It tests the choices you make as the person using MCP servers: where to configure them, what to expose, how tools get picked, and how failures come back.

First, the picture

Claude Code or your agent (client) Jira MCP server Tools: create_issue, update_issue Resources: open issues summary Orders MCP server (in-house) Tools: lookup_order, process_refund Resources: database schema All servers' tools are found at start-up and usable together.

The four questions at a glance

#QuestionUsual answerPractise
1Where does the server config go?Shared: .mcp.json. Personal: ~/.claude.json. Secrets: ${VAR}MCP: where and what
2Tool or resource?A resource shows a catalog so the agent stops exploring. A tool does something or fetches one thingMCP: where and what
3Why did Claude pick the wrong tool?Fix the description first, then scope tools per agent, then use tool_choice for orderTool picker, Too many tools, tool_choice, Built-in tools
4A tool failed. What does it return?If the tool ran, isError: true with a category and whether to retry. If the call was broken, a protocol errorTool failure

A yellow row in each map is the runner-up: the case where a different answer wins because one fact in the question changes.

Mind map 1: Where does the server config go?

What you seeWhere the fault isFix
The whole team needs the same serverIt's in one person's personal configProject-level .mcp.json, committed to the repo
One developer is trying out a serverIt was put in the shared file, so everyone gets itUser-level ~/.claude.json
A token is written into .mcp.jsonA secret is committed to gitUse env var expansion: ${GITHUB_TOKEN}
The team needs Jira or GitHubThey plan to build their own serverUse the community MCP server
The system is in-house, like an orders databaseNo community server exists for itBuild a custom server. This is when building wins
Rule: shared config goes in .mcp.json, personal config in ~/.claude.json, and secrets always go through ${VAR}. Use a community server for standard systems.

Guide 2.4: “Scoping: project-level .mcp.json (shared) vs user-level ~/.claude.json (personal/experimental). Env var expansion in .mcp.json (e.g., ${GITHUB_TOKEN}) — no committed secrets.”

Example. A team wants every developer who clones the repo to get the GitHub MCP server. Each developer has their own GitHub token. What should they do?
  1. Each developer adds the server to ~/.claude.json
  2. Add it to .mcp.json in the repo, with the token as ${GITHUB_TOKEN}
  3. Add it to .mcp.json with a shared team token
  4. Describe the server in CLAUDE.md
B. It's shared through git, and no secret is committed. A works, but nobody gets it automatically. C commits a secret. D is just text, so no server is set up.

Mind map 2: Tool or resource?

What you seeWhere the fault isFix
The agent makes many search or list calls just to find out what existsNothing shows it the catalog up frontExpose the catalog as an MCP resource (issue summaries, a doc tree, a schema)
The agent must change something, like creating an issue or refunding an order—That's a tool
The agent needs one specific record, like order 4411—That's a tool with an input
The content changes on every request or needs parametersA fixed resource would be out of date or too bigA tool wins here
Rule: a resource answers “what's available?” so the agent stops exploring. A tool does something or fetches exactly one thing.

Guide 2.4: “MCP resources expose content catalogs (issue summaries, doc hierarchies, schemas) to reduce exploratory tool calls.”

Example. Before writing any query, an agent calls list_tables and then describe_table about a dozen times. How do you cut those calls?
  1. Merge the two tools into one explore_db tool
  2. Tell the agent to call describe_table less often
  3. Expose the database schema as an MCP resource
  4. Raise the agent's tool-call limit
C. The agent sees the schema up front, so it doesn't have to explore. A works, but it's still exploration. B asks the agent to guess. D allows more calls instead of fewer.

Mind map 3: Why did Claude pick the wrong tool?

What you seeWhere the fault isFix
Two similar tools get mixed upThe descriptions are short or overlapRewrite the descriptions: input format, example queries, edge cases, and when not to use it
One generic tool like analyze gets used wronglyThe tool does too many jobsSplit it into purpose-specific tools with clear inputs and outputs
Claude uses Grep instead of your more capable MCP toolThe MCP tool's description is weakImprove the MCP tool's description. Don't remove Grep
One agent has 18 tools and picks badlyIt has too many toolsSplit the work so each agent has the 4 or 5 tools for its role
A subagent uses a tool outside its jobTools aren't scoped to rolesScope tools per agent. A few shared tools are fine if every role uses them often
A tool must run first, like extract_metadatatool_choice is auto, so Claude may skip itForce it: {"type":"tool","name":"extract_metadata"}
Any tool call is fine, but never a plain-text replyauto allows texttool_choice: "any", not a forced tool
Rule: if the wrong tool is picked, fix the description first. If there are too many tools, scope them per agent. If the order matters, use tool_choice. An extra line in the system prompt is usually the weaker answer.

Guide 2.1: “Tool descriptions are the primary mechanism LLMs use for selection.” Guide 2.3: “Too many tools (18 vs 4-5) degrades selection reliability.” Guide 2.4: “Enhance MCP tool descriptions so the agent doesn't prefer built-ins (Grep) over more capable MCP tools.”

Example. A support agent has get_customer (“Gets customer”) and lookup_account (“Looks up account”). It often calls the wrong one. Best fix?
  1. Add a routing table to the system prompt
  2. Rewrite both descriptions with what each returns, example inputs, and when to use the other one
  3. Merge them into one customer_tool
  4. Force get_customer with tool_choice
B. The descriptions are the cause, so fix them. A works, but it patches around the descriptions. C hides the difference. D breaks every account lookup.

Mind map 4: A tool failed. What does it return?

First ask one question: did the tool run? If the call itself was broken, the tool never ran. If the tool ran and its work failed, the result says so.

What you seeWhere the fault isFix
Unknown tool name, or a required argument is missingThe call is broken, so the tool never ranA JSON-RPC protocol error
The tool ran, but the downstream API failed (even with a 404)The work failedA result with isError: true, which Claude sees and can act on
The agent keeps retrying something that can never workThe error only says “Operation failed”Add errorCategory, isRetryable and a plain-language description
A timeoutTransient errorisRetryable: true
A policy blocks the refundBusiness-rule errorretriable: false and a customer-friendly explanation
“No results”, but the system was actually downAn access failure looks like an empty resultMake these two clearly different responses
A subagent sends every small error to the coordinatorIt escalates too earlyRecover locally. Send up only unresolved errors, with the partial results
Rule: a broken call is a protocol error. A tool that ran and failed returns isError: true, with the category and whether retrying makes sense. Never send a bare “failed”.

Guide 2.2: “MCP isError flag communicates failures back to the agent. Distinguish transient (timeouts), validation (bad input), business (policy), permission errors. Uniform ‘Operation failed’ prevents good recovery decisions.”

Example. A refund tool calls the payments API, which rejects the refund because it is over the policy limit. What should the tool return?
  1. A JSON-RPC protocol error
  2. isError: true with the text “Operation failed”
  3. isError: true, category business, retriable: false, and a short reason the agent can tell the customer
  4. An empty success result
C. The tool ran, so it's isError, and the details stop pointless retries. A is for broken calls. B works, but the agent can't decide what to do. D hides the failure.

Two traps that catch people

1 · Protocol error or isError?

Ask “did the tool run?” A 404 from the downstream API sounds like a protocol error, but the tool ran, so it's isError. Newer versions of the MCP spec suggest isError for bad arguments too. On the exam, follow the guide. More on Exam vs 2026, item 5.

2 · “Add it to the system prompt”

When one option fixes the description, the tool scope or tool_choice, it usually beats an option that only adds a prompt instruction. A prompt asks Claude to comply, while a fix to the tool setup removes the cause.

How to study this

  1. Read one mind map, then open its Trap Gym page. Change the setting and watch what happens.
  2. Do the D2 practice questions. For each one, say which row of which map decides it.
  3. Print this page and cover the Fix column. Can you fill it in from memory?

Related: Domain 2 hub · MCP skeletons · Exam vs 2026 · all traps · CCA-F start here