What is the Model Context Protocol (MCP)? A plain-English guide
MCP is the open standard that lets AI models use external tools and data. Here's what it is, how it works, and why it matters for building agents.
The Model Context Protocol (MCP) is an open standard that lets an AI model talk to external tools and data sources — your email, a database, a search engine, a payment API — through one consistent interface. If REST gave web apps a common way to talk to servers, MCP gives AI agents a common way to talk to capabilities.
Why MCP exists
Before MCP, every team wired tools into their agent by hand: bespoke function definitions, custom auth, glue code per integration. Swap the model and you rewrote it. MCP standardizes the contract so any model can discover and call any MCP-compatible tool, and any tool can be reused across apps.
The core idea
An MCP server exposes a set of tools (actions the model can take), resources (data it can read), and prompts. An MCP client — usually inside an AI app like Claude Code or Cursor — connects to that server, lists what's available, and lets the model call it. The model sees a clean menu of capabilities instead of raw APIs.
A tool call looks roughly like: the model asks "what can I do here?", gets a list with input schemas, picks one, and calls it with arguments. The server runs it and returns a result the model can reason over.
Local vs. remote servers
MCP servers come in two flavours:
- Local servers run as a process on your machine (started with
npx,uvx, or Docker). Great for personal use, but there's no URL to share and nothing multi-tenant. - Remote servers are hosted at an HTTPS endpoint (Streamable HTTP or SSE). These are the ones you can connect from anywhere — and the ones a platform can manage for you.
We go deeper on the trade-offs in Remote vs. local MCP servers.
Where the pain shows up
MCP solves the protocol, but production agents still need a lot around it: managing OAuth and secrets for each server, keeping tool definitions out of the prompt, adding memory, enforcing guardrails on destructive calls, and tracing every invocation. That connective tissue is exactly what Vortic provides — connect remote MCP servers, bundle them into a box, and reach them all from one endpoint in Claude, Cursor, or your own app.
Getting started
The fastest way to feel it: browse the MCP server directory, pick a server like Context7 or Linear, and connect it. In a minute you'll have a single endpoint your agent can use to discover and run that server's tools — no per-tool wiring.