Model Context Protocol (MCP): a practical intro

MCP (Model Context Protocol) is a way for an LLM app (the client) to connect to external capabilities exposed by one or more MCP servers—things like reading a repo, querying a database, calling an internal API, or running safe automations.

The value is standardization: instead of building custom integrations per tool, your client speaks MCP to many servers.

The moving parts

• MCP Client: your app/IDE/agent that decides when to call tools.
• MCP Server: exposes capabilities (often “tools”) in a consistent format.
• Tools / Resources: the actual operations and data sources behind the server.

Typical request flow

When MCP is a good fit

• You want one client to talk to many internal tools safely.
• You want tool schemas and consistent calling patterns.
• You want to keep credentials + network access on the server side, not inside the LLM client.

Practical tips

• Keep tools small and composable (one job per tool).
• Return structured output (JSON-ish) when possible.
• Treat servers as "capability boundaries" (auth, audit logs, rate limits).

---

Example: App MCP Server Implementation

Let's walk through a concrete example—building an MCP server that exposes a tool to query the App API info endpoint. The server uses stdio transport, runs in Docker, and integrates with Cursor's agent window.

Architecture

Files to Create

1. MCP Server (mcp/server.py)

Main MCP server implementation using the mcp Python SDK:

from mcp.server import Server
from mcp.server.stdio import stdio_server
import requests

app = Server("app-mcp-server")

@app.tool()
async def get_app_info() -> dict:
    """Query the App API info endpoint."""
    response = requests.get(
        "https://api-dv2.app-usw2.dev.aligntech.com/app/info"
    )
    response.raise_for_status()
    return response.json()

async def main():
    async with stdio_server() as (read_stream, write_stream):
        await app.run(read_stream, write_stream)

if __name__ == "__main__":
    import asyncio
    asyncio.run(main())

2. Dependencies (mcp/requirements.txt)

mcp>=1.0.0
requests>=2.28.0

3. Dockerfile (mcp/Dockerfile)

FROM python:3.11-slim

WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

COPY server.py .

CMD ["python", "server.py"]

4. Docker Compose (optional) (mcp/docker-compose.yml)

For easier container management during development:

version: "3.8"
services:
  app-mcp:
    build: .
    image: app-mcp-server
    stdin_open: true
    tty: true

5. README (mcp/README.md)

Documentation with Cursor configuration instructions.

Cursor Configuration

After building the Docker image, add to .cursor/mcp.json:

{
  "mcpServers": {
    "app": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "app-mcp-server"]
    }
  }
}

Build and Run Steps

1. Build Docker image: docker build -t app-mcp-server ./mcp
2. Restart Cursor to pick up the MCP configuration
3. Use the agent window to ask about App API status

Tool Specification

The MCP server exposes one tool:

Tool	Description	Parameters
get_app_info	Query the App API info endpoint	None