Something weird is happening in the AI ecosystem.
A protocol called MCP (Model Context Protocol) went from zero to 859 integrations on Apify Store in 3 months. GitHub repos with "MCP" in the name doubled since January. And nobody's talking about it.
Let me explain why this matters — and why it might change how you build apps.
What MCP Actually Is (30-Second Version)
Traditional API:
Your App → HTTP Request → API Server → JSON Response → Your App parses it
MCP:
AI Agent → "I need weather data" → MCP Server → Structured response → AI uses it directly
The key difference: the AI decides when and how to use the tool. You don't write API call logic. You describe what tools are available, and the AI figures out the rest.
Why This Is a Big Deal
1. It kills boilerplate
Traditional approach to building an app that searches the web, scrapes data, and writes a summary:
# 50+ lines: auth, requests, error handling, parsing, retrying...
response = requests.get(url, headers=headers, timeout=30)
if response.status_code != 200:
# retry logic
pass
data = response.json()
results = data['items'][:10]
# ... more parsing
MCP approach:
# The AI calls tools as needed. You just describe them.
tools = [
{"name": "web_search", "description": "Search the web"},
{"name": "scrape_url", "description": "Get page content"},
]
# That's it. The AI handles the rest.
2. It's composable
An MCP server that searches the web + an MCP server that reads PDFs + an MCP server that writes to a database = an AI that can research any topic and store the results. No integration code needed.
3. It's model-agnostic
MCP works with Claude, GPT, Llama, Gemini — any model that supports tool calling. Write your MCP server once, use it with any AI.
The Numbers Don't Lie
I analyzed the Apify Store (the largest MCP marketplace):
- 859 MCP-compatible actors (up from ~200 in December 2025)
- Top MCP actors: 24 users each (for new actors, this is significant)
- Categories: web scraping, research, data enrichment, monitoring
- Growth rate: ~100 new MCP actors per month
GitHub tells a similar story:
- "MCP server" repos: 2,000+ (doubled in Q1 2026)
- Most starred MCP repos: 500-2,000 stars
- Languages: Python (60%), TypeScript (35%), Go (5%)
How to Build Your First MCP Server (5 Minutes)
from mcp.server import Server
from mcp.types import Tool, TextContent
import requests
server = Server("my-web-tool")
@server.tool("fetch_url")
async def fetch_url(url: str) -> list[TextContent]:
"""Fetch and return the text content of a URL."""
resp = requests.get(url, timeout=10)
return [TextContent(type="text", text=resp.text[:5000])]
if __name__ == "__main__":
server.run()
Add this to your claude_desktop_config.json, and Claude can now read any URL you ask about.
What I Think Happens Next
- MCP marketplaces will emerge (Apify is already one)
- Every SaaS will ship an MCP server alongside their REST API
- "MCP engineer" becomes a job title by end of 2026
- The API economy gets a competitor: the MCP economy
Am I Wrong?
I might be overhyping this. MCP could plateau. APIs might be "good enough" for most use cases.
What do you think? Is MCP a real shift, or just another protocol that'll be forgotten in a year?
I'd love to hear from anyone who's built or used MCP servers in production.
I built a Claude MCP Cookbook with 15 ready-to-use examples. And here's a curated list of 77+ web scraping tools including MCP-compatible ones.
