All 43 questions
What is MCP, and how does it connect AI agents to tools and live data?
Asked in 8 of 181 interviews · last asked Sep 18 · Tooling · Mid-level
Show answer
MCP, the Model Context Protocol, is an open standard that connects AI apps to tools and data sources. The app uses an MCP client to connect to servers that expose tools, data resources, and prompt templates. Agents can use these connections to query live data or take actions without a custom connection format for every service.
- The client discovers available tools and sends tool calls to the server.
- Servers can connect to databases, files, APIs, and business systems.
- Data is live only if the connected source provides current data.
- MCP standardizes communication; apps still need permissions and approval rules for sensitive actions.
What the interviewer checks: The interviewer checks whether you understand MCP’s client-server design, practical uses, and limits.
Other ways it was asked
- “What do you know about MCP, the Model Context Protocol, for tool use?”
- “What is the Model Context Protocol (MCP), and what problem does it solve?”
- “What is the Model Context Protocol (MCP), and what are its use cases?”
What is an MCP server?
Asked in 5 of 181 interviews · last asked Oct 9 · Concept · Junior
Show answer
An MCP server is a program that exposes tools, data, or reusable prompts through the Model Context Protocol. An AI application connects to it using an MCP client. This lets the application access external systems in a standard way, such as reading files or calling an API.
- MCP stands for Model Context Protocol.
- Servers can expose tools, resources, and prompts; they do not need to provide all three.
- Servers can run locally or remotely.
- The AI application controls access and tool use; the server does not replace the model.
What the interviewer checks: The interviewer is checking whether you understand how MCP connects AI applications to external tools and data.
Other ways it was asked
- “What does an MCP server do?”
- “What is an MCP server, or Model Context Protocol server?”
- “What are MCP servers, and how are they used in LLM applications?”
How does an agent select tools, who executes them, and how do results become a final answer?
Asked in 3 of 181 interviews · last asked Oct 21 · Concept · Mid-level
Show answer
The app sends the model the user request and tool descriptions, including input schemas. The model chooses a tool based on the task and returns its name and arguments. The app checks the request, runs the tool, and sends its result back to the model. The model uses the result to answer or request more tools.
- Tool calling means requesting a function call; the model does not execute the function itself.
- The app validates arguments and checks permissions before execution.
- Tool results or errors go back with the matching call identifier.
- The model may answer without tools, or use several calls before answering.
What the interviewer checks: The interviewer checks whether you understand the tool loop and separate model decisions from app execution.
Other ways it was asked
- “How do agents choose and invoke tools internally?”
- “How do tools work in an LLM agent, and who is responsible for calling them?”
- “What is tool calling/function calling? Walk us through the four steps from when your code sends a request to when the user gets a final answer.”
How do you choose between building custom agent tools and using existing APIs or native integrations?
Asked in 2 of 181 interviews · last asked Oct 14 · Design · Senior
Show answer
I use existing APIs or native integrations when they meet the task, security, and reliability needs. I build custom tools when I need missing features, special business rules, or tighter control. I compare setup effort, ongoing maintenance, cost, and vendor dependence. Often, a small custom wrapper around an existing API is enough.
- Check required features, permissions, rate limits, and error handling before choosing.
- Native integrations can save setup work, but may limit control or customization.
- Custom tools offer control but require testing, monitoring, and maintenance.
- Keep tool permissions narrow and require approval for sensitive actions.
What the interviewer checks: The interviewer is checking whether you balance task fit, security, and long-term maintenance instead of always building from scratch.
Other ways it was asked
- “How would you implement tool calling for an AI agent: build custom tools and invoke them through an API, or use native integrations such as Google Calendar?”
- “When building an AI agent, how do you decide whether to use existing tools or write custom tools?”
What are external tools in an AI agent, and how does MCP relate to them?
Asked in 2 of 181 interviews · last asked Sep 8 · Concept · Mid-level
Show answer
External tools are functions or services an AI agent can call, such as search, databases, or code runners. MCP stands for Model Context Protocol. It gives AI applications a standard way to discover and call tools exposed by MCP servers. I would use it to reuse tool connections across compatible applications; direct API calls are also an option.
- Tools let agents get current data or take actions beyond generating text.
- An MCP client connects to a server that exposes tools and describes their inputs.
- MCP standardizes the connection; it does not replace the underlying APIs or decide which tool to use.
- Tool access still needs permissions, input checks, and approval for risky actions.
What the interviewer checks: The interviewer checks whether you understand tools, MCP’s role, and when a standard connection is useful.
Other ways it was asked
- “What is MCP, and when would you use it to connect an AI agent to tools?”
How would you build an MCP server, and what components would it include?
Asked in 2 of 181 interviews · last asked Sep 8 · Design · Mid-level
Show answer
I would build an MCP server with an official SDK, tool definitions, and handlers that call application services. The AI application would host an MCP client that discovers and calls those tools. I would use stdio for a local process or Streamable HTTP for a remote server. I would add input validation, access checks, error handling, and logs.
- The SDK handles protocol messages, initialization, and capability negotiation.
- Expose three tools: search documents, fetch a document, and create a support ticket.
- Resources provide readable context; prompts provide reusable message templates. Both are optional.
- Keep credentials on the server; require user approval for sensitive changes in the client application.
What the interviewer checks: The interviewer checks whether you understand MCP’s client-server design and can expose useful capabilities safely.
Other ways it was asked
- “How would you architect an application using an MCP server, including the client-side components?”
- “How would you design an MCP server for an application, and what three tools or capabilities would it expose?”
How does an MCP agent select a forecasting model?
Asked once · last asked Aug 24 · Concept · Senior
When is MCP useful, and when would you use it?
Asked once · last asked Feb 5 · Concept · Mid-level
How would you retrieve the dynamic data from the live database, including through MCP?
Asked once · last asked Feb 5 · Scenario · Mid-level
Can you explain what plugins are in Semantic Kernel?
Asked once · last asked Apr 24 · Concept · Mid-level
What kinds of tools can an AI agent call?
Asked once · last asked Apr 29 · Concept · Mid-level
How would you integrate LinkedIn or a similar platform with an LLM?
Asked once · last asked Jul 17 · Scenario · Mid-level
In a scenario with 10,000+ tools, how would you handle tool selection when you cannot list all tool descriptions in the prompt?
Asked once · last asked Jul 17 · Design · Senior
What are the benefits of using MCP compared with LangGraph?
Asked once · last asked Jul 20 · Concept · Mid-level
When might you choose not to use the Model Context Protocol (MCP) with Figma in a software development workflow?
Asked once · last asked Jul 28 · Concept · Mid-level
How do you invoke a planner agent through an API?
Asked once · last asked Jul 31 · Scenario · Mid-level
What advantages does an MCP server offer over a conventional server or regular backend logic?
Asked once · last asked Aug 4 · Concept · Mid-level
How does calling an API differ from interacting with an MCP server?
Asked once · last asked Aug 4 · Concept · Mid-level
How would you decide which tool an AI agent should call in an access-assistance system?
Asked once · last asked Aug 7 · Design · Senior
Which agent-to-agent protocols can simplify secure communication without requiring teams to build their own protocol?
Asked once · last asked Aug 20 · Concept · Senior
How would you integrate an AI agent with external tools and systems using APIs, servers, or other methods, and assign responsibility for those integrations?
Asked once · last asked Aug 22 · Design · Senior
Other ways it was asked
- “How does it reach or integrate with external tools and systems? Does it use APIs, servers, or some other method, and who is responsible for these integrations?”
How would you decide how many tools to integrate into an AI agent system and which to expose to the agent, considering the effects on performance and complexity?
Asked once · last asked Aug 22 · Design · Senior
Other ways it was asked
- “How many tools are integrated within the system? Does having many tools affect performance or complexity, and how do you determine which tools to expose to the agent?”
How would you architect an MCP server to let an agent safely query a financial database without breaking the production system?
Asked once · last asked Aug 26 · Design · Senior
Which transport mechanism would you use for communication between an MCP client and server?
Asked once · last asked Aug 26 · Concept · Mid-level
How does function calling enable an LLM to act as an agent?
Asked once · last asked Sep 4 · Concept · Mid-level
What problem does MCP solve compared with wiring Claude directly to each API?
Asked once · last asked Sep 8 · Concept · Mid-level
In an AI agent's tool wrapper or integration layer, how would you classify and surface retryable and non-retryable errors, rate limits, authentication failures, and validation errors so the model can choose another action?
Asked once · last asked Sep 8 · Design · Senior
How would you build an MCP server using TypeScript, including initialization, capabilities, tools, and prompts?
Asked once · last asked Sep 21 · Write code · Senior
How do you handle asynchronous operations and timeouts between user calls through an MCP server and database responses?
Asked once · last asked Sep 21 · Scenario · Senior
Why use a separate intent classifier instead of letting the LLM decide tool invocation through native function calling?
Asked once · last asked Sep 22 · Concept · Senior
How would an agent plan tool calls for a query comparing a semiconductor company's stock movements with the earnings growth of several technology companies?
Asked once · last asked Sep 22 · Scenario · Senior
Other ways it was asked
- “How would an agent plan tool calls for a query comparing NVIDIA's stock movements with the earnings growth of OpenAI, Microsoft, or Qualcomm?”
How would you connect existing Microsoft platforms and layer AI and automation on top?
Asked once · last asked Oct 1 · Design · Senior
How can AI agents be integrated with MCP servers or skills?
Asked once · last asked Oct 8 · Concept · Mid-level
What is MCP, and why would you use it instead of function calling?
Asked once · last asked Oct 9 · Concept · Mid-level
What is the difference between an API and a tool in the context of an AI agent?
Asked once · last asked Oct 19 · Concept · Mid-level
Other ways it was asked
- “What is a tool in the context of an AI agent?”
What tools and information should an AI agent be able to access to complete a task?
Asked once · last asked Oct 21 · Design · Mid-level
How does MCP compare to A2A, and when would you reach for each?
Asked once · last asked Oct 21 · Concept · Senior
What is the difference between MCP and function calling? Are they competing or complementary?
Asked once · last asked Oct 21 · Concept · Senior
If no MCP server is available for a tool such as Azure, Google Cloud, or Databricks, how could you connect it to Claude Code?
Asked once · last asked Oct 22 · Scenario · Senior
How long would it take to build an MCP server for integration with SAPO?
Asked once · last asked Jul 23 · Scenario · Mid-level
What are automated tool agents?
Asked once · last asked Jul 25 · Concept · Junior
What role does the FastMCP component play in an MCP-based system?
Asked once · last asked Aug 24 · Concept · Mid-level
What are MCP SDKs, and how do you use them to build an MCP server?
Asked once · last asked Sep 21 · Concept · Mid-level
More Agentic AI interview questions
- All Agentic AI interview questions
- Agentic AI Basics and Use Case Interview Questions
- AI Agent Architecture and Multi-agent Interview Questions
- AI Agent Memory, Context Management, and RAG Interview Questions
- AI Agent System Design Interview Questions
- AI Agent Production Deployment and Reliability Interview Questions
- AI Agent Security, Guardrails, and Governance Interview Questions
How this list is made
Every question here was asked by a real interviewer in a real job interview that a candidate ran MeetAssist in. After the call, the interviewer's questions are pulled from the transcript. Practice sessions, tests, exams and anything that was not a job interview are dropped.
Questions are rewritten so they make sense on their own: names of people, employers and clients are removed, and speech-to-text mistakes are fixed. Small talk, logistics and questions about one candidate's own résumé are left out. Rephrasings of the same question are merged, and the count is the number of different interviews it came up in.
The interviews come from candidates who use MeetAssist, so they lean towards software, data and AI roles: the shares describe these interviews, not every Agentic AI interview. Answers to the most asked questions are written and checked by an editor; a "common mistake" is shown only when reviews of real answers in these interviews show it, with the count.
181 interviews, 116 candidates, Nov 2025 to Oct 2026. Updated every month with new interviews.
Or add the Chrome extension