Mindket
English
English简体中文
Sign inSign up
StoreMCPSkillsCalculatorLearnPriceBlog
Discover
Research308Writing59Business95Legal13Finance & accounting33Marketing & SEO172Developer1,379Productivity382Education81Companions14
Help and support
Language
English简体中文
Tool guideMCP by AnthropicOpen source

MCP for beginners: what it is and how to use it

Mindket editors · Published · 6 min read

MCP, the Model Context Protocol, is an open standard that lets an AI app such as Claude, ChatGPT, VS Code or Cursor use outside systems. Each system sits behind an MCP server, a program on your computer or at a web address that offers tools the AI can call and data it can read. Anthropic created MCP in 2024, and the protocol is free to use.

Good for
  • Connect AI to your files
  • Call tools from chat
  • Query databases and APIs
  • One server for many apps

At a glance

  • MCP is a standard, not an app. You use it through an AI app that supports it, or through a test client such as MCP Inspector.
  • Check whether a server runs on your computer or at a web address, because that decides who receives your requests.
  • To watch a server work with no AI app involved, launch MCP Inspector and call one tool.
  • A local server runs with your account's permissions, so read its start command before you add it.
All tools
On this page
  1. YouTube tutorials
  2. What MCP is
  3. What MCP is good for
  4. When to use MCP, and when to pick another tool
  5. How to use MCP, step by step
  6. An example to try
  7. Tips for using MCP well
  8. Learn more about MCP
  9. Frequently asked questions
  10. Sources

What MCP is

MCP, short for Model Context Protocol, is an open standard for connecting AI apps to the systems where your files, data and tools live. Anthropic open-sourced it on 25 November 2024, and OpenAI and Google DeepMind have since adopted it. In December 2025 Anthropic donated it to the Agentic AI Foundation, a fund under the Linux Foundation.

You don't install MCP the way you install an app. It is a set of rules that two programs follow. The AI app you chat in, such as Claude, ChatGPT, VS Code or Cursor, is the host. Each outside system sits behind an MCP server. That is a small program, on your own computer or at a web address, that offers tools the AI can call, data it can read and prompts you can pick. The protocol is free, though some servers connect to services that need your own account or API key.

What MCP is good for

  • Reads and changes your files: the reference filesystem server lets an app list, read, copy and create files, but only in folders you allow.
  • Turns a design into code: Claude Code can read a Figma mockup's components, design tokens and layout rules through MCP and build from them.
  • Checks your calendar and notes: per MCP's own introduction, an agent can read Google Calendar and Notion through their servers.
  • Queries a database: in the architecture guide's example, tools run the queries, a resource holds the schema and a prompt shows how to call them.
  • Adds your own tool to every AI app: write one small server in Python or TypeScript, and any app that speaks MCP can call it.
  • Lets n8n workflows talk to AI: n8n has a node that lets an agent call an MCP server, and a trigger that turns n8n itself into one.

When to use MCP, and when to pick another tool

Your jobUse MCP?Or try
Give an AI app access to your files, a database or a web serviceYes, through an MCP serverClaude Code or Cursor as the app that uses it
Add your own tool to several AI apps at onceYes, write one serverThe Python lesson builds one
Run steps on a schedule or when a form is filled inNon8n or Zapier
Chat, write or research with no outside systemsNoChatGPT, Claude or Gemini
Run a model on your own computerNo, MCP does not run modelsOllama or LM Studio
Teach an AI app a procedure written as instructionsNo, that is a skill's jobAgent Skills, in our guide

How to use MCP, step by step

MCP has no sign-up page. The quickest way to see it work is MCP Inspector, the official test client, pointed at a published server. The six pictures call one tool on the filesystem server, so you need no AI app and no code. Put a few files in a folder first, then install the server in another folder with npm install @modelcontextprotocol/server-filesystem.

  1. In a terminal, check that node --version prints 22.23.4, 24.22.1 or 26.12.1, or a later release on the same line, because the Inspector asks for a patched Node. Then type npx -y @modelcontextprotocol/inspector, followed by node, the full path to the server's dist\index.js and the folder of files. The terminal prints a web address with a token in it. The token in the picture belongs to a launch that has ended, so yours will differ.

    A terminal that shows the node version check, then the Inspector launch command and the launcher output below it. The command and the output are both ringed in gold. The output holds the web address with its token.

  2. Your browser opens that address by itself. If no tab opens, copy the address from the terminal. Do not type it from memory, because the token changes at every launch. On the Servers screen, switch on the toggle on your server's card. The card shows the command the Inspector will run.

    The Servers screen of MCP Inspector. A Read-only session note sits above a card named node, marked STDIO and Disconnected, with its start command. The connect toggle on the card is ringed in gold.

  3. Once connected, the page shows the server's own name. Open the Tools tab. The list heading gives the number of tools, 16 in this case, and the Protocol column on the right has already logged the tools/list request that fetched them.

    The Tools tab, ringed in gold, of an Inspector connected to secure-filesystem-server. The list is headed Tools (16) and starts with Append File, Copy File and Create Directory. A Protocol column on the right shows three logged messages.

  4. Type list in the search box and select List Directory. Fill the path field with the folder you gave the server, then select Execute Tool.

    The Tools tab with List Directory selected. Its path field holds a test folder, and the Execute Tool button is ringed in gold.

  5. The answer appears under Results, one line for each file or folder, marked [FILE] or [DIR]. An AI app does the same when it calls a tool, except that the model fills in the form for you.

    The Tools tab after a call. The Results panel, ringed in gold, shows notes.txt as a file, recipes as a folder and shopping.md as a file. The Protocol column shows a tools/call message.

  6. Select the Protocol tab, then Expand on the tools/call card. The Parameters box holds what the client sent and the Response box holds what the server sent back, both as JSON-RPC.

    The Protocol column with the card for the tools/call request to list_directory opened, its arrow control ringed in gold. Its Parameters box holds the tool name and a path argument. Its Response box holds a jsonrpc 2.0 result with the file list as text.

Screenshots of MCP Inspector 2.14.0 on Windows, taken on 6 November 2026 with the published filesystem server. The terminal picture is a frame drawn around the real command and its output. That launch ran with the browser auto-open turned off, so its output has no Opening browser... line.

The tutorials at the top of this page show MCP at work inside AI apps. The MCP Inspector lesson repeats these steps with every command.

An example to try

Start the Inspector on the reference filesystem server. The server's own start command follows the Inspector's name, and the last part is a folder the server may read. Replace it with the full path of a folder that holds a few test files.

# The server's own start command goes after the Inspector's name
npx -y @modelcontextprotocol/inspector npx @modelcontextprotocol/server-filesystem /path/to/a/test/folder
  • A good start shows a server card on the Servers screen. Switch it on and the page lists the server's name and its tools.
  • Calling List Directory returns your files, each marked [FILE] or [DIR].
  • The Protocol tab holds a tools/call card with the exact request and response.
  • If nothing connects, open Console. It shows what a local server wrote to stderr.

Tips for using MCP well

  • Read a server's start command before you add it to any app. A local server runs under your account, so it can do what you can do by hand.
  • Check whether a server is local or remote first. A remote one receives your requests at its web address, and a local one starts on your own computer.
  • Use absolute paths in config files. Apps start servers from their own working directory, so a relative path such as ./data can fail.
  • Use restricted API keys for servers that need one, as Cursor's docs advise.
  • Quit and reopen Claude Desktop after you edit its config. Closing the window is not enough.

Learn more about MCP

  1. 1MCP course: build, test and connect MCP servers · Lesson 1 of 4Build an MCP server in Python with the official SDKBuild a Python MCP server with the official SDK: install it with uv, write two tools, run it over stdio, then check it in MCP Inspector and Claude Desktop.8 min
  2. 3MCP course: build, test and connect MCP servers · Lesson 3 of 4MCP Inspector: test and debug an MCP serverLearn MCP Inspector: launch it on a local or remote server, call tools in the browser, read the protocol log and run the same checks from the command line.11 min
  3. 4MCP course: build, test and connect MCP servers · Lesson 4 of 4Connect an MCP server to Claude Desktop, Cursor and VS CodeConnect an MCP server to Claude Desktop, Cursor and VS Code: each app's config file, its JSON format, how to reload it and where to look when nothing shows up.8 min
  4. 2MCP course: build, test and connect MCP servers · Lesson 2 of 4Build an MCP server in TypeScript with the official SDKBuild a TypeScript MCP server with the official SDK: set up the project, register a tool with a Zod schema, serve it over stdio and call it from the Inspector.7 min

Frequently asked questions

Is MCP free to use?

Yes. The protocol is free, with its code and specification contributions under the Apache License 2.0. Some servers connect to services that need your own account or API key, and those services set their own terms.

Do I need to code to use MCP?

No. Adding a server to an app is a settings change, not programming. Code comes in only when you write a server of your own, and the official SDKs cover several programming languages.

Is an MCP server the same as an API?

No, although many servers call an API underneath. An API is one service's own interface, and every service designs its own. An MCP server wraps what it offers in a shared format, so an app that speaks MCP can list and call its tools without code written for that service.

Which AI apps work with MCP?

Claude, ChatGPT, Visual Studio Code and Cursor are among them, according to MCP's own introduction. Each keeps its own config file: Claude Desktop reads claude_desktop_config.json, Cursor reads .cursor/mcp.json and VS Code reads .vscode/mcp.json.

Is it safe to add any MCP server?

No. A local server runs with your account's permissions, so add one only after you have read its start command. The MCP specification adds that a host must get your explicit consent before it calls a tool. Our guide to MCP server safety lists what to check first.

How is MCP different from Agent Skills?

A skill is a folder of instructions, and an MCP server is a live connection to a system. They can sit side by side, and our Agent Skills guide explains where each one fits.

How do I test an MCP server I wrote?

Run MCP Inspector with your server's start command after its name, such as npx @modelcontextprotocol/inspector node build/index.js. Then connect, call a tool and read the messages in the Protocol tab.

Sources

  1. Wikipedia: Model Context Protocol (read November 6, 2026)https://en.wikipedia.org/wiki/Model_Context_Protocol
  2. Model Context Protocol: What is MCP? (read November 3, 2026)https://modelcontextprotocol.io/docs/getting-started/intro
  3. MCP docs: Architecture overview (read November 2, 2026)https://modelcontextprotocol.io/docs/learn/architecture
  4. MCP docs: Understanding MCP servers (read November 2, 2026)https://modelcontextprotocol.io/docs/learn/server-concepts
  5. MCP specification: Tools (read November 2, 2026)https://modelcontextprotocol.io/specification/latest/server/tools
  6. Anthropic: Introducing the Model Context Protocol (read November 2, 2026)https://www.anthropic.com/news/model-context-protocol
  7. Anthropic: Donating the Model Context Protocol and establishing the Agentic AI Foundation (read November 3, 2026)https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation
  8. GitHub: modelcontextprotocol/servers (read November 2, 2026)https://github.com/modelcontextprotocol/servers
  9. Claude on X: Claude Code and Figma through MCP, 23 September 2025 (read November 3, 2026)https://x.com/claudeai/status/1970541285615264071
  10. n8n Docs: MCP servers (read November 2, 2026)https://docs.n8n.io/build/integrate-ai/mcp-servers
  11. n8n Docs: MCP Server Trigger node (read November 2, 2026)https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-langchain.mcptrigger
  12. MCP docs: MCP Inspector (read November 4, 2026)https://modelcontextprotocol.io/docs/tools/inspector
  13. MCP docs: Inspector web client (read November 4, 2026)https://modelcontextprotocol.io/docs/2026-07-28/tools/inspector/web
  14. MCP Inspector: README (read November 4, 2026)https://github.com/modelcontextprotocol/inspector
  15. MCP docs: Connect to local MCP servers (read November 4, 2026)https://modelcontextprotocol.io/docs/2026-07-28/develop/connect-local-servers
  16. MCP docs: Debugging (read November 4, 2026)https://modelcontextprotocol.io/docs/2026-07-28/tools/debugging
  17. Cursor docs: Model Context Protocol (read November 4, 2026)https://cursor.com/docs/mcp
  18. VS Code docs: MCP configuration reference (read November 4, 2026)https://code.visualstudio.com/docs/agents/reference/mcp-configuration

Share this lesson

Build your first MCP server

Lesson one builds a small weather server in Python with the official SDK.

Start the lesson

This lesson as Markdown · All tools · All courses