MCP client

#How an MCP client works

A host can use several servers at once. It runs one client for each server, so each connection stays separate. When a client connects, it asks its server which tools it offers, and it shows those tools to the model. When the model requests one, the client sends the call to the server and passes the result back to the model.

Apps differ in how they show tool calls and whether they ask you before a call runs. Check your app's settings to see what it asks.

#Why it matters for AI agents

The client decides which servers an agent can reach. Adding a server to a client's configuration gives the model new tools. Look at the list of connected servers in your app to see what the agent can call.

#Wield and MCP clients

Wield's installer, npx -y wield-mcp init, adds Wield to Claude Desktop, Claude Code, Cursor and Codex. It asks before it changes each config file, and it makes a backup first. Claude Desktop can also install Wield from a one-click MCPB bundle. Other MCP clients work with the standard stdio configuration. See AI apps that work with Wield.

#Related terms

• MCP server: the program the client connects to.
• stdio transport: the local connection most desktop clients use with local servers.
• MCPB bundle: a package that installs a server into a desktop client.

#FAQ

#What is the difference between an MCP host and an MCP client?

The host is the AI app you use, such as Claude Desktop. The client is the part of the host that manages one connection to one MCP server. A host usually contains several clients, one for each server it uses.

#Which AI apps are MCP clients?

Claude Desktop, Claude Code, Cursor and Codex are the apps Wield documents, and wield-mcp init can set them up. Other apps can also act as MCP clients. Any client that supports the standard stdio configuration can run wield-mcp. Check each app's documentation for its MCP support.