What is MCP?

What is MCP?

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:

TypeWhat it doesExample
ToolsFunctions the AI can executeRun a PromQL query
ResourcesData the AI can readFetch a Grafana dashboard
PromptsTemplated workflowsPRD 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

ServerPurpose
dot-aiKubernetes operations, shared prompts, PRD workflows
GrafanaDashboard queries, alerts, incidents
PrometheusPromQL queries, metric exploration
Context7Up-to-date documentation for any library
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)”

Resources:

Last updated on