MCP Servers

January 30, 2026

MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems. Think of it like USB-C for AI — a standardized way to connect agents to data sources, tools, and workflows.

The Core Problem

AI agents need to interact with the world: databases, APIs, file systems, browsers, calendars. Before MCP, every integration was custom — different protocols, auth methods, and data formats for each connection.

MCP standardizes this. With a single protocol, agents can connect to:

Data sources: Local files, databases, cloud storage
Tools: Search engines, calculators, code execution
Workflows: Specialized prompts, automation pipelines

How It Works

MCP uses a client-server architecture:

┌─────────────┐      ┌─────────────┐      ┌─────────────┐
│   Claude    │      │ MCP Server  │      │  External   │
│   (Client)  │◄────►│  (Bridge)   │◄────►│   System    │
└─────────────┘      └─────────────┘      └─────────────┘

The MCP Server acts as a bridge between the AI client and external systems. Servers expose:

Resources: Data the agent can read
Tools: Actions the agent can take
Prompts: Specialized instruction templates

What MCP Enables

Real-world use cases:

Agents accessing Google Calendar and Notion for personalized assistance
Claude Code generating web apps using Figma designs
Enterprise chatbots connecting to multiple databases
AI models creating 3D designs in Blender

As one summary explained:

It's an open standard from Anthropic that connects AI assistants to data sources like repositories and tools via secure, standardized servers. This simplifies integrations, breaks down data silos, and boosts AI relevance.

Industry Adoption

MCP has become the universal standard. As noted on Twitter:

Amazon, Microsoft, Google, OpenAI, and Anthropic are all backing it. 10,000+ published servers. 97 million monthly SDK downloads.

Enterprise infrastructure providers are fully committed:

AWS — Deployment support
Cloudflare — Hosts MCP servers
Microsoft Azure — Full integration
Google Cloud — Managed server offerings

The protocol is now stewarded by the Agentic AI Foundation under the Linux Foundation.

MCP Apps: Interactive UI

Anthropic recently extended MCP with MCP Apps — servers that can render interactive UIs directly in the AI client.

Example workflow with MCP Apps:

Claude drafts a FigJam rollout with error-rate gates
Checks Amplitude split test data (staging +40% over control)
Syncs Asana tasks automatically
Drafts Slack update for team review — within the chat

Connectors launched for Slack, Figma, Asana, Amplitude, Hex, and more.

Popular MCP Servers

A few servers worth knowing:

Context7 — Live code documentation for AI coding agents
Docker MCP — Manages multiple MCPs with sandbox security
Shadcn Registry — Easy access to UI component registries
Google Cloud MCPs — Maps, BigQuery, Compute integration
Notion/Obsidian — Workspace content automation
Supabase — Backend automation from DB to SQL

Security Considerations

MCP introduces new attack surfaces. Recent research identified vulnerabilities in MCP servers:

The primary risk stems from the combination of two factors: LLMs can be tricked via prompt injection, and MCP servers blindly trust the LLM's decisions.

Key vulnerabilities found:

CVE-2025-68143: Unrestricted git_init in attacker-chosen directories
CVE-2025-68145: Path validation bypass for file deletion
CVE-2025-68144: Argument injection for reading arbitrary files

Best practices:

Implement strict input validation
Use sandboxing (Docker containers)
Apply least-privilege principles
Deploy prompt injection safeguards

Building MCP Servers

Quick start with the official SDK:

import { McpServer } from "@modelcontextprotocol/sdk/server";

const server = new McpServer({
  name: "my-server",
  version: "1.0.0",
});

// Define a tool
server.addTool({
  name: "get_weather",
  description: "Get weather for a city",
  inputSchema: { type: "object", properties: { city: { type: "string" } } },
  handler: async ({ city }) => {
    // Fetch weather data
    return { temperature: 72, condition: "sunny" };
  },
});

MCP vs Skills

MCP and Agent Skills are complementary:

Purpose — MCP: connect to systems. Skills: provide instructions.
Type — MCP: protocol + servers. Skills: Markdown + files.
Focus — MCP: actions & data. Skills: knowledge & workflows.
Activation — MCP: always available. Skills: loaded on demand.

Use MCP when the agent needs to do something or access something. Use Skills when the agent needs to know how to do something.

Sources

Model Context Protocol — Official documentation
Agentic AI Foundation — Linux Foundation governance
MCP vulnerability research — Security considerations