What is MCP?

2 min read

MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external tools and data sources. The simplest way to think about it: MCPs are APIs for AI agents.

Just like REST APIs let applications talk to services, MCP lets AI assistants talk to tools — databases, monitoring systems, file systems, documentation, anything. Same concept, different consumer.

The Problem MCP Solves

Without MCP, every AI integration is custom-built. Want Claude to query Prometheus? Build a connector. Access Grafana dashboards? Another connector. Search documentation? Another one. Each integration is bespoke, fragile, and duplicated across every AI tool.

MCP flips this: build once, use everywhere. A Prometheus MCP server works with Claude Code, ChatGPT, VS Code Copilot, and any other MCP-compatible client. The server exposes capabilities; the client consumes them.

How It Works

MCP uses a client-server architecture:

    graph LR
    subgraph "AI Application"
        Client[MCP Client]
    end

    subgraph "MCP Servers"
        S1[Prometheus]
        S2[Grafana]
        S3[Filesystem]
    end

    Client -->|MCP Protocol| S1
    Client -->|MCP Protocol| S2
    Client -->|MCP Protocol| S3

Servers expose three types of capabilities:

Type	What it does	Example
Tools	Functions the AI can execute	Run a PromQL query
Resources	Data the AI can read	Fetch a Grafana dashboard
Prompts	Templated workflows	PRD creation workflow

Why It Matters

MCP became the industry standard fast. Anthropic created it in November 2024; by 2025, OpenAI and Google DeepMind adopted it. In late 2025, Anthropic donated MCP to the Linux Foundation’s Agentic AI Foundation.

The ecosystem now has thousands of community-built servers, official SDKs for all major languages, and 97M+ monthly downloads.

MCP Servers I Use

Server	Purpose
dot-ai	Kubernetes operations, shared prompts, PRD workflows
Grafana	Dashboard queries, alerts, incidents
Prometheus	PromQL queries, metric exploration
Context7	Up-to-date documentation for any library
Arc DevTools	Browser control via Chrome DevTools Protocol

MCP servers consume context window. Be strategic — keep global MCPs minimal and use per-repo configs for project-specific tools. See AI Tips.

Learn More

“The Missing Link: How MCP Servers Supercharge Your AI Coding Assistant”

“MCP Servers Explained: Why Most Are Useless (And How to Fix It)”

Related: Commands vs MCP vs Skills — How MCP compares to other agent extension mechanisms.

Resources:

MCP Specification — Official protocol docs
MCP GitHub — SDKs and reference servers
Awesome MCP Servers — Community server collection
Anthropic MCP Announcement — Original introduction

Last updated on 16 March 2026

Tags: ai mcp claude-code