Model Context Protocol (MCP)

#How it works

An MCP server describes its tools with a name, a description and a JSON Schema for the inputs. The AI app shows those tools to the model. When the model decides to use one, the app sends a tools/call request to the server and passes the result back to the model. The messages are JSON-RPC 2.0.

Servers connect in one of two ways:

• stdio: the app starts the server as a local process and talks to it over standard input and output. Nothing listens on the network.
• Streamable HTTP: the server runs elsewhere and the app connects to it over HTTP.

#Why it matters for agents

Before MCP, every AI app needed its own plugin format. With MCP, one server works in any client that speaks the protocol. That's why a single tool can be installed in Claude Desktop, Claude Code, Cursor and Codex with the same few lines of config.

#MCP and Wield

Wield is an MCP server. wield-mcp runs locally over stdio and gives your agent a few tools: list_sites, find_recipe, run, request_recipe and status. With run, the agent can use a website through a recipe that runs in your own browser. See what the tools do and the sites that work today.

#FAQ

#Is MCP only for Claude?

No. It's an open specification, and many clients support it, including Cursor, Codex, VS Code and Zed. See AI apps that work with Wield.

#Is an MCP server the same as an API?

No. An API is whatever a service exposes. An MCP server is an adapter that presents tools to an AI app in one standard shape. It may call an API, run a local program, or, in Wield's case, run requests in your browser.