The 5 Best MCP Servers for Claude Desktop (And How to Install Them in 10 Minutes)

Claude Desktop is one of the most powerful AI tools available today. But out of the box, it has a fundamental limitation: it can’t interact with your local environment.

It can’t read a file from your computer. It can’t fetch a web page without you pasting the content. It forgets everything the moment you close the window.

MCP servers fix this. The Model Context Protocol (MCP) is an open standard that lets Claude connect to external tools and data sources — your filesystem, databases, the web, and more.

In this article, I’ll walk through the 5 MCP servers I consider essential for any serious Claude Desktop user, explain what each one does, and show you how to get all five running in under 10 minutes.

What Is MCP? (Quick Primer)

MCP (Model Context Protocol) is an open standard developed by Anthropic that allows AI models like Claude to call external tools. Think of it like browser extensions — except for your AI assistant.

Each MCP server exposes a set of “tools” that Claude can invoke during a conversation. When you ask Claude to read a file, it calls your File Manager server. When you ask it to fetch a URL, it calls your Web Scraper server. The servers run locally on your machine and communicate with Claude Desktop via stdio — you configure them once in claude_desktop_config.json and they’re available in every conversation.

The 5 Essential MCP Servers

1. File Manager — Read, Write & Search Local Files

The most immediately useful server. Once installed, you can say things like “Summarize all the text files in my Projects folder”, “Save this as a Markdown file on my Desktop”, or “Search my Documents folder for anything mentioning invoice”. The server is sandboxed to a root directory you configure — Claude can only access files within that boundary, which keeps things safe.

Tools: file_read, file_write, file_list, file_search, file_delete

2. Web Scraper — Any URL as Clean Markdown

How many times have you copied an article into Claude? This server eliminates that. Point Claude at any public URL and it fetches the page, strips navigation and ads, and returns clean Markdown that Claude can reason about directly. It also blocks internal and private addresses, so it won’t reach into your local network.

Tools: web_fetch, web_fetch_multi (fetch up to 5 URLs concurrently)

3. SQLite Query Tool — Natural Language Database Queries

If you have any data in a SQLite database, this server is transformative. You ask questions in plain English; Claude writes the SQL, executes it, and returns readable results. It runs in read-only mode by default, so there’s no risk of accidentally modifying your data.

Tools: sqlite_query, sqlite_execute, sqlite_schema, sqlite_sample

4. PDF Exporter — Professional Output in One Command

Claude produces great content; getting it out of the chat window in a professional format is awkward. Say “export this as a PDF” and you get a properly formatted document — custom typography, page numbers, code styling, tables — saved straight to your Desktop. (Uses WeasyPrint for headless rendering and falls back to clean HTML if WeasyPrint’s system libraries aren’t installed.)

Tools: pdf_export, pdf_check_deps

5. Persistent Notes — Memory That Survives Sessions

Claude has no memory between conversations. The Notes server gives Claude a persistent notepad you control — save ideas, meeting summaries, project context, then retrieve them in future sessions by title or tag.

Tools: note_create, note_get, note_update, note_delete, note_search

How to Install All Five in 10 Minutes

I’ve packaged all five servers into the MCP Starter Kit — with a one-click installer for macOS, Linux, and Windows.

What the installer does:

1. Finds a suitable Python (3.10+) — and if none is installed, offers to install it for you
2. Installs the required packages (mcp, httpx, pydantic)
3. Validates each server file
4. Patches your claude_desktop_config.json automatically — merging safely so your existing servers are preserved, with a backup

macOS / Linux:

cd mcp-starter-kit/scripts
chmod +x install.sh
./install.sh

Windows (PowerShell):

cd mcp-starter-kit\scripts
.\install.ps1

Restart Claude Desktop. Done. And when you want it gone, a matching uninstall script removes everything cleanly and leaves your other servers untouched.

👉 Get the MCP Starter Kit on Gumroad — $29

Building Your Own MCP Servers

Want to go further? The kit is MIT licensed and every server is clean, commented Python — designed as a learning resource as much as a tool. The pattern for a new server is straightforward:

from mcp.server.fastmcp import FastMCP
from pydantic import BaseModel, Field

mcp = FastMCP("my_server_mcp")

class MyInput(BaseModel):
    query: str = Field(..., description="Your input")

@mcp.tool(name="my_tool")
async def my_tool(params: MyInput) -> str:
    """What this tool does."""
    return f"Result for: {params.query}"

if __name__ == "__main__":
    mcp.run()

Common Questions

Do I need coding experience to install the kit? No. The installer handles everything. If you can open a terminal and run one command, you’re set.

Does this work with Claude.ai (the web version)? No — MCP servers only work with Claude Desktop. Make sure you have the desktop app installed.

Can I modify the servers for my own use case? Yes — MIT license, full source included. The code is designed to be extended.

👉 MCP Starter Kit — $29 on Gumroad

Built in Vienna. Questions? richard@kuhn.or.at