Skip to content
PortLev PortLev Learn

What Is Mcp

What Is MCP?

Section titled “What Is MCP?”

MCP (Model Context Protocol) is the standard that lets Claude connect to external systems. It defines how AI models and tools communicate — what format requests take, how results are returned and how capabilities are declared.

The Problem MCP Solves

Section titled “The Problem MCP Solves”

Without MCP, every AI tool integration is bespoke: you write custom code to connect Claude to Gmail, write more custom code to connect it to Notion, more for GitHub, more for Slack. Each integration is one-off, hard to share and hard to maintain.

MCP standardizes the interface. An MCP server exposes tools in a common format. Claude Code (and any other MCP-compatible client) can use any MCP server without custom integration code.

Architecture

Section titled “Architecture”
Claude Code (MCP Client)
        ↕ MCP Protocol (JSON-RPC over stdio or HTTP)
MCP Server (e.g., Gmail server)
        ↕ Native API
Gmail API

The MCP server translates between the MCP protocol and the native API. You never write that translation code — you just install the server.

What MCP Servers Expose

Section titled “What MCP Servers Expose”

Every MCP server declares:

Tools — Actions Claude can take

send_email(to, subject, body)
create_issue(repo, title, body, labels)
query_database(sql)

Resources — Data Claude can read

file://path/to/document.pdf
gmail://inbox/message-id
notion://page/page-id

Prompts — Reusable prompt templates

"Summarize the last 10 emails from this sender"
"Create a weekly status report from these Notion pages"

Transports

Section titled “Transports”

MCP servers communicate over two transports:

stdio (most common): Claude Code launches the server as a child process, communicates via stdin/stdout. Simplest setup — just install the server and configure the path.

HTTP/SSE (for remote servers): The server runs on a URL, Claude connects over the network. Used for shared servers, team-wide MCP, or serverless deployment.

Security Model

Section titled “Security Model”

MCP servers only run when Claude Code is active. They don’t run in the background. Claude requests tools from them on-demand, and the MCP server executes on your machine with your credentials.

Key points:

  • MCP servers run with the permissions of the user who starts them
  • Credentials (API keys, OAuth tokens) stay on your machine — they’re never sent to Anthropic
  • You can restrict which MCP servers are available per project via .claude/settings.json

The Registry

Section titled “The Registry”

Thousands of MCP servers are publicly available. Browse them at:

Next: Installing MCP Servers