MCP Role-Play and Quiz

Summary

🧭 What this page covers

MCP with Python closed the implementation pages. This final module page consolidates the material with an interview-style role-play and a short quiz.

If you can explain where MCP sits, when to use tools vs resources vs prompts, and how host/client/server responsibilities differ, you understand the module at an engineering level.

Visual Recap Before Quiz

1) Concept map

2) Runtime lifecycle

3) Decision table

Decision point	Prefer this	Reason
Read-only knowledge access	Resource	Lower-risk content access, clearer semantics
Action with side effects	Tool	Explicit execution contract and validation
Repeatable instruction workflow	Prompt	Reusable template discoverable by hosts
One-host local integration	stdio transport	Fast and simple operational model
Multi-host shared service	Remote HTTP/SSE	Centralized governance and reuse

Senior Tech vs Dev Role-Play

Senior Tech: Explain MCP to an architecture review board in one minute.

Dev: MCP is a standard protocol for exposing AI-adjacent capabilities from servers to AI hosts. It lets hosts discover tools, resources, and prompts rather than hard-coding every integration. It does not replace APIs or model function calling; it standardises the application boundary around them.

Senior Tech: What is the most common design mistake when teams first adopt MCP?

Dev: Treating everything as a tool. Read-oriented content should usually be a resource, and reusable workflow framing should often be a prompt. If everything becomes a callable action, the model gets a larger, riskier execution surface and hosts lose useful structure.

Senior Tech: Where should authorization live?

Dev: In both places, but for different reasons. The host curates what the model can see for a given scenario. The server enforces the final authorization boundary so a malicious or buggy host cannot bypass policy.

Senior Tech: Local stdio or remote HTTP server?

Dev: Start with local stdio if the capability serves one host and low-latency local execution is fine. Move to remote hosting when multiple hosts need the same capability or when centralized auth and operations justify the distributed systems overhead.

Senior Tech: How would you explain MCP vs function calling to a product manager?

Dev: Function calling is how the model asks the app to do something during a conversation. MCP is how the app discovers and connects to reusable external capabilities. One is a model interaction pattern; the other is an integration protocol.

Quiz

1. What problem does MCP primarily solve?

   Details

   Answer

   It standardises how AI hosts discover and invoke reusable tools, resources, and prompts from external servers.

2. Which MCP primitive should you use for a read-only policy document?

   Details

   Answer

   A resource.

3. Which primitive is best for create_ticket?

   Details

   Answer

   A tool, because it performs an action and may have side effects.

4. Who owns approval and policy decisions in an MCP setup?

   Details

   Answer

   Primarily the host, with final authorization enforced by the server.

5. Why is stdio often the best first transport?

   Details

   Answer

   It is simple, local, and avoids distributed systems complexity when only one host needs the capability.

6. Does MCP replace your existing REST APIs and databases?

   Details

   Answer

   No. MCP usually sits on top of them as an AI-facing integration layer.

7. What is the risk of modelling resources as tools?

   Details

   Answer

   The action surface becomes larger and less clear, and hosts lose the semantic distinction between content access and execution.

8. Why should a .NET MCP server stay thin?

   Details

   Answer

   To reuse existing domain services rather than duplicate business logic in a new layer.

9. What lifecycle step happens before invoking a tool?

   Details

   Answer

   Initialization and discovery of server capabilities.

10. What is the cleanest short explanation of MCP vs function calling?

Details

Answer

Function calling is how the model expresses an action request; MCP is how applications expose and discover reusable capabilities around that model.

Common Pitfalls

Pitfall	What goes wrong	Prevention
Memorising terms without architecture understanding	Can repeat definitions but cannot design a system	Practice explaining host/client/server responsibilities
Ignoring governance in examples	Designs look clean but unsafe	Include auth, approval, and audit in every architecture discussion
Thinking MCP is only for coding tools	Misses enterprise use cases	Apply the pattern to search, policy, operations, and workflows

References and Next Steps

• Previous: MCP with Python →
• Next module: Providers, Models, and Selection →
• Specification: modelcontextprotocol.io