annsa
Log inTry Annsa

Glossary · MCP client

What is an MCP client?

An MCP client is the AI application that connects to MCP servers. It is the side that asks.

In practice

Claude Code, Cursor, Claude, ChatGPT and several editors can act as clients. They discover what a server offers and call its tools during a session.

“Client” and “host” are sometimes used for the same thing. When people separate them, the host is the app you use and the client is its connection to 1 server.

The client isn't the model. The model chooses a tool. The client makes the call and returns the result.

Why it matters for product builders

Client support decides who can reach a server. If it works in only 1 editor, everyone else is excluded.

Questions people ask.

Q1

What's the difference between an MCP client and an MCP server?

An MCP client is the part of an AI app that connects to one MCP server. The app it lives in, like Claude Code or Cursor, is the host; the server provides the tools and data. When Claude Code reads your ranked list from Annsa, its client connects to Annsa's MCP server.
Q2

Which apps are MCP clients?

Many AI tools you build with include one: Claude Code, Cursor, Codex and Grok Build, as well as ChatGPT, Windsurf, VS Code, Figma Make, Lovable and Bolt. Each connects to remote MCP servers by URL, and each can connect to Annsa at https://app.annsa.ai/mcp.
Q3

How many tools can an MCP server give a client?

The protocol doesn't set a number, but each tool's description takes up room in the agent's context when it's loaded, so fewer, clearly named tools tend to work better. Annsa's server has eight, grouped by what they do: three read, and five write.

Last reviewed 2026-08-21

Put it to work with your agents.

Begin with one agent and the work that matters next.

Try AnnsaSee pricing