AI : LLM : MCP Servers

MCP Quick Guide

A concise overview of the Model Context Protocol and related tooling. Internal anchors let you jump between sections. External links point to primary sources.

1. MCP and MCP servers - overview

MCP stands for Model Context Protocol. It is an open protocol that lets AI models access external resources through a consistent and controlled interface. An MCP server is the component that exposes data and actions as tools. The AI model acts as an MCP client and calls those tools to read or act on data. This pattern avoids ad hoc connectors per tool and helps keep access predictable and auditable.

For background, see the External resources or visit Wikipedia - Model Context Protocol and the Vercel overview of MCP.

Back to top

2. FastMCP - quick intro

FastMCP is a Python framework that makes it easy to build MCP servers. You define tools and resources in Python and FastMCP handles the protocol details. It is useful for rapid prototypes and local integrations. For production, you still apply your own auth, rate limits, and logging.

Project home: gofastmcp.com and github.com/jlowin/fastmcp.

Back to top

3. What you can expose to an AI via MCP

Anything reachable via a library, API, or CLI can be wrapped as an MCP tool. Common categories include:

  • File system access - read, write, search local files
  • Databases - SQL and NoSQL queries
  • APIs - REST, GraphQL, JSON RPC
  • Web search - controlled queries with guardrails
  • Email - draft, send, or search through provider APIs
  • Calendar - create or read events
  • Contacts - org directories and address books
  • Chat apps - Slack, Teams, Discord, WhatsApp gateways
  • Cloud storage - Google Drive, OneDrive, Dropbox
  • Spreadsheets - Excel and Google Sheets read or update
  • Docs and notes - Confluence, Notion, Wikis
  • Issue trackers - Jira, GitHub, GitLab issues and PRs
  • DevOps tools - Jenkins, GitHub Actions, Kubernetes triggers
  • Version control - Git fetch, diffs, commit history
  • Monitoring and logs - Splunk, Datadog, CloudWatch
  • Finance and ERP - invoices and records lookup
  • CRM - Salesforce and HubSpot data
  • AI tools - vision, speech, embeddings, transcription
  • Custom scripts - Python, Node, or shell utilities
  • Hardware and IoT - sensors, cameras, printers
  • Security tools - SIEM alerts and IAM checks
  • Knowledge bases - SharePoint and internal wikis
  • E-commerce - Shopify or payments gateway calls
  • Analytics - run BI queries and dashboards

Back to top

4. Hardware company use case

A computer hardware company can run an MCP server that connects its AI assistant to sales analytics, supply chain databases, and support logs. The AI asks for monthly sales by product, correlates them with inventory shortages, and checks warranty claims for specific component batches. Leaders can ask questions like: which regions show declining laptop sales linked to GPU shortages, and do rising tickets for Model X align with returns. These insights help optimize production, forecast demand, and improve reliability.

Back to top

5. One line definition

MCP servers let me safely integrate data sources or APIs with my LLM, so the model can access and use that data through a controlled protocol.

Back to top

Back to top

7. Minimal FastMCP server example

from fastmcp import FastMCP

mcp = FastMCP("Demo")

@mcp.tool
def add(a: int, b: int) -> int:
    return a + b

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

This registers a single tool named add that returns a + b. Expand with auth, logging, and rate limits for production.

Back to top