Quiz 2: MCP in Practice

Test your understanding of building, configuring, and deploying MCP systems. Choose the best answer.

Question 1: FastMCP Server Building

选项

• A. Use FastMCP decorators (@mcp.tool()) to define functions as MCP tools
• B. Write raw JSON-RPC messages for each tool
• C. Gradio is the only way to build MCP servers
• D. MCP servers must be written in JavaScript

答案解析

• A（正确答案） — Correct! FastMCP infers types from function signatures and generates schemas automatically.
• B — Possible but tedious. FastMCP handles the protocol for you.
• C — Gradio is convenient, but FastMCP is more flexible for complex logic.
• D — MCP is language-agnostic. Python, JavaScript, Go, Rust all work.

Question 2: Type Hints and Docstrings

选项

• A. Type hints and docstrings are essential for MCP tool schemas
• B. Type hints are optional; MCP infers types from usage
• C. Only Gradio MCP requires type hints
• D. Docstrings are just for human readers

答案解析

• A（正确答案） — Exactly! Type hints generate schemas, docstrings provide descriptions. Always include both.
• B — Type hints are essential. Without them, MCP can't generate proper schemas.
• C — All MCP servers benefit from type hints. Clients need to know parameter types.
• D — Docstrings are used by MCP to generate tool descriptions and argument docs.

Question 3: Tools vs. Resources vs. Prompts

选项

• A. Tools: actions | Resources: read-only data | Prompts: instructions
• B. Tools and Resources are interchangeable
• C. Resources are for UI only, not for agents
• D. Prompts are the same as Tools

答案解析

• A（正确答案） — Perfect! Each serves a different purpose in agent workflows.
• B — No. Tools are called by agents; Resources are data agents read.
• C — Resources are designed for agents! They provide context agents can access.
• D — No. Prompts are instruction templates; Tools are callable functions.

Question 4: Gradio MCP Integration

选项

• A. Use demo.launch(mcp_server=True) to enable MCP automatically
• B. Gradio and MCP are separate; you need both libraries
• C. You need to manually implement the MCP protocol in Gradio
• D. Gradio MCP only works with simple functions

答案解析

• A（正确答案） — Correct! One parameter enables both web UI and MCP endpoint.
• B — Gradio includes MCP support with gradio[mcp]. No separate library needed.
• C — Gradio handles it automatically. Just write functions with type hints.
• D — Gradio MCP works with complex functions, multiple parameters, and advanced types.

Question 5: Configuring Agents to Use MCP Servers

选项

• A. Configure servers via CLI commands (like claude mcp add) or config files, depending on the agent
• B. Agents automatically discover all available MCP servers
• C. You need a special GUI tool to configure MCP servers
• D. Each agent needs its own custom MCP SDK integration

答案解析

• A（正确答案） — Correct! Claude Code uses CLI commands; other agents use config files. Remote Streamable HTTP servers can often be added without restarts.
• B — No. You must explicitly configure which servers each agent can access.
• C — Nope. Just edit JSON config files directly.
• D — False! MCP is a standard protocol. Agents are built as MCP clients already.

Question 6: Deploying to Hugging Face Spaces

选项

• A. Create a Space, upload app.py with mcp_server=True, add requirements.txt
• B. Spaces doesn't support MCP yet
• C. You need to configure network ports and manage server processes
• D. MCP servers on Spaces only work with private access

答案解析

• A（正确答案） — Perfect! Spaces auto-builds and exposes the MCP endpoint.
• B — Spaces fully supports MCP via Gradio. It's the recommended deployment target.
• C — Spaces handles all of this. Just push your app code.
• D — Spaces can be public or private. MCP works with both.

Question 7: Transport Mechanisms

选项

• A. Use stdio for local servers, Streamable HTTP for remote deployment
• B. Streamable HTTP is always better because it's 'network-native'
• C. Once deployed, you must switch from stdio to Streamable HTTP
• D. Transport is internal to MCP; agents don't see the difference

答案解析

• A（正确答案） — Exactly! Stdio is fast for subprocess communication. Streamable HTTP is the current standard for remote MCP connections.
• B — Streamable HTTP has network overhead. Use stdio locally for speed.
• C — You choose transport at configuration time. No switching mid-deployment.
• D — True! Agents use the same tool interface regardless of transport.

Question 8: MCP Security

选项

• A. Use environment variables for secrets; never hardcode them in config
• B. Hardcode API tokens in config if it's 'private'
• C. Use relative paths in server configuration
• D. There's no security consideration for MCP configuration

答案解析

• A（正确答案） — Excellent! Env vars keep secrets out of version control.
• B — Even private configs can be exposed. Always use environment variables.
• C — Use absolute paths! Relative paths break depending on working directory.
• D — Configuration files can contain sensitive info. Protect them carefully.

Summary

How many did you get right?

• 7-8 correct: You're ready to build and deploy real MCP systems!
• 5-6 correct: Good understanding. Review configuration and deployment sections.
• 3-4 correct: Review the building and integration sections.
• 0-2 correct: Go back through the unit and review the core concepts.

Key Takeaways

• FastMCP is the simplest way to build servers with decorators
• Type hints and docstrings are essential for MCP tool schemas
• Gradio with mcp_server=True creates UI + MCP automatically
• Configure servers in agent config files (or use MCPClient in code)
• Use stdio locally, Streamable HTTP for remote deployment
• Always use environment variables for secrets
• Deploy to Hugging Face Spaces for easy sharing

You've mastered Unit 2! You can now build, deploy, and configure MCP servers for any agent.