引導式測驗 · 中等 等級
使用 MCP Inspector 測試 MCP 伺服器測驗
測試 Inspector 的客戶端角色、stdio 和 HTTP 傳輸、舊版與 2026 MCP 發現、直接工具測試、驗證和模型使用限制。
問題 1 的 8
What does MCP Inspector primarily let a developer do?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- What does MCP Inspector primarily let a developer do?
- Connect directly to an MCP server and test its tools, resources and prompts without a live model
- Train a new language model on tool-call examples
- Automatically write tool descriptions for a server
- Deploy an MCP server to a cloud host
- Why might a model never call a tool even though it works correctly in Inspector?
- Inspector always disables tools after ten calls
- The tool's description or parameter documentation may be too unclear for the model to know when to use it
- MCP servers expire after 24 hours
- The model cannot read JSON-RPC responses
- Over what transport does Inspector typically communicate with a local server process?
- Bluetooth pairing
- A dedicated UDP broadcast
- stdio, exchanging newline-delimited JSON over stdin and stdout
- FTP file transfer
- How does capability discovery differ between legacy MCP and revision 2026-07-28?
- Both revisions require an LLM API key
- The newer revision removes tools/list
- Legacy uses HTML forms rather than JSON-RPC
- Legacy uses initialize; modern 2026-07-28 omits that handshake and requires server/discover support, which clients may call optionally
- A tool call in Inspector that returns a JSON-RPC error object is best understood as what?
- Diagnostic evidence about the request or handler to inspect alongside the method, parameters, and server logs
- A crash requiring a full server rewrite
- Evidence that MCP itself is broken
- A sign the model hallucinated the tool
- What should a team understand about HTTP transports when testing MCP servers?
- All MCP HTTP traffic must use the retired legacy SSE transport
- Streamable HTTP can return JSON or use SSE; older HTTP+SSE remains a compatibility path
- Remote servers cannot use HTTP
- HTTP transport replaces every server tool
- Does Inspector itself require a live LLM provider API key to list and call server tools?
- Yes, every MCP client calls a language model
- Yes, unless the server uses stdio
- No, but the MCP server may still require its own credentials or authorization
- No, and remote MCP servers can never authenticate
- Which of these is a genuine limitation of testing solely with MCP Inspector?
- It cannot show request or response payloads
- It cannot connect to stdio servers
- It requires the server to be written in Python
- It cannot reveal whether a model will actually choose to invoke a given tool in real use