What is an MCP server?

An MCP server is the USB port for AI: the model thinks, MCP lets it act — read files, query a database, drive a browser. Three questions: should you install one, which one, and how.

The one-line version

Without MCP a model can only chat; with MCP it can act. It is not an in-app plugin but a shared protocol: one server works in Claude Code, Cursor and Cherry Studio.

Plugins vs function calling vs MCP: a plugin extends one app; function calling is the model's ability to call a function; MCP is the shared socket — write a server once, reuse it in many clients.

① Should you install one?

• Only chatting in a web page → no.
• Using Claude Code / Cursor / Cherry Studio for code or documents → two or three are enough; do not start with ten.
• You need the AI to read files or a database on your machine → install the matching server.

② Which one (five starter categories)

• Files: filesystem — let the AI read and write a folder.
• Web: fetch / browser — open pages and extract text.
• GitHub: repos, issues, pull requests.
• Database: sqlite / postgres — query tables in plain language.
• Documents: local document search; start small.

Our catalogue has 7,801 MCP and skill entries; see what others install this week: MCP hub →

③ How to add one (three clients)

Claude Code (CLI):

claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem /path/to/folder

Cursor (config at ~/.cursor/mcp.json):

{"mcpServers": {"filesystem": {"command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/folder"]}}}

Cherry Studio: Settings → MCP servers → Add → paste the JSON above.

Command names and package names follow each project's README; restart the client and check the tool list.

Related

MCP hub (7,801 entries) → · Claude Code hub → · All AI coding tools →