11MCP: Connecting Agents to Everything

The problem MCP solves

Before MCP, every AI app needed custom code for every integration: one connector for Google Drive in app A, a different one in app B. The Model Context Protocol is an open standard — think of it as a USB-C port for AI. Build or install a connector once as an MCP server, and any MCP-compatible app can use it: Claude, Claude Code, Hermes Agent, n8n, VS Code, Cursor and many more.

What an MCP server can offer

Servers run either locally (the app launches them as a process — “stdio” transport) or remotely over HTTP, which is how hosted connectors for services like Notion or Asana work.

Connect a ready-made server

The official filesystem server lets an AI app read and write files — but only inside folders you allow. Most desktop hosts accept a config like this (the file location differs per app; check its docs):

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/you/ClientProjects"
      ]
    }
  }
}

With Claude Code, the same thing is one command:

claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem ~/ClientProjects

Restart the app and ask: “List the files in my ClientProjects folder and summarize the newest proposal.”

Build your own MCP server (Python)

Expose your own business data as tools any agent can use. This server gives agents read access to invoices.

1. Install the SDK

   pip install "mcp[cli]"

2. Write the server

   from mcp.server.fastmcp import FastMCP
   import csv
   
   mcp = FastMCP("invoices")
   
   def _load():
       with open("invoices.csv", newline="") as f:
           return list(csv.DictReader(f))
   
   @mcp.tool()
   def get_invoice(invoice_id: str) -> dict:
       """Get one invoice by its ID: client, amount in PHP, due date and status."""
       for row in _load():
           if row["id"] == invoice_id:
               return row
       return {"error": f"Invoice {invoice_id} not found"}
   
   @mcp.tool()
   def unpaid_total() -> str:
       """Total amount (PHP) of all unpaid invoices."""
       total = sum(float(r["amount_php"]) for r in _load() if r["status"] == "unpaid")
       return f"PHP {total:,.2f}"
   
   if __name__ == "__main__":
       mcp.run()          # stdio transport — the host app launches this process

3. Test it with the MCP Inspector

   mcp dev invoice_server.py

   A browser inspector opens where you can call each tool manually.

4. Register it in your AI app

   {
     "mcpServers": {
       "invoices": {
         "command": "python",
         "args": ["/full/path/to/invoice_server.py"]
       }
     }
   }

Security: MCP servers are code you run