What is MCP

MCP — the Model Context Protocol — is an open standard that lets AI agents discover and invoke capabilities at runtime. An MCP tool, once published, can be listed and called by any JSON-RPC 2.0-compatible agent without custom integration. For the short version, see What is an MCP Server.

Why it matters

Instead of building a one-off integration for every system, MCP gives capabilities a shared shape — like a standard plug that fits any socket. For an enterprise, that means new integrations arrive in a consistent, reviewable form rather than as bespoke code nobody else understands.

How it works

An MCP tool receives a JSON-RPC 2.0 request and returns a JSON-RPC 2.0 response. A request looks like this:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/list",
  "params": {}
}

And a response:

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": { }
}

You can write MCP tools in Python, JavaScript, or Lua.

Supported methods

Method	Purpose
initialize	Handshake (required)
ping	Health check
tools/list	Return available tools and their schemas
tools/call	Execute a specific tool
resources/list / resources/read	List or read resources
prompts/list / prompts/get	List or get prompts
completion/complete	Handle completion requests

A minimum implementation needs initialize, tools/list, and tools/call.

In the enterprise

Which MCP servers are available, and who can add them, is governed like any other connector or tool. See Connector governance & security. For full code examples in each language, see Writing Code.