Agentic AI Notebook
Model Context Protocol
Phase 8Module 4 of 25

MCP Server

The value of MCP is on the server side: one wrapper around Jira, Datadog, or your internal API.

The server is a power adapter. Your service has a weird plug. MCP is the standard outlet every client already knows.

Visual Workflows

Start here — scroll inside each diagram frame to explore, then use + / to zoom up to 200% if needed.

Overview

100%
Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

100%
Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

One tool call

100%
Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

The LLM never talks to Jira directly. It talks to the MCP server.

Key Takeaways

  • 1.A server wraps a real system and speaks MCP to any client.
  • 2.It advertises tools with name, description, and JSON Schema.
  • 3.call_tool runs your code and returns text (or images) as content.
  • 4.Write the server once — Cursor and your agent both consume it.
  • 5.Keep tool descriptions honest — the LLM chooses tools from those sentences.

Learn elsewhere

  • Full build steps — Build MCP Server
  • Resources vs tools — next two modules

Real Example

Scenario

A Jira server exposes create_ticket, search_issues, and add_comment. Cursor and a support agent both use it without knowing Jira's REST shape.

What you would do

In Model Context Protocol, apply MCP Server to this scenario: A Jira server exposes create_ticket, search_issues, and add_comment. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Commands

Commands to Remember

  • list_tools advertises JSON Schema
  • call_tool runs your function
  • Return TextContent, do not print to stdout
  • stdio servers must keep stdout clean for JSON-RPC

Cheat Sheet

Quick recap

quick ref
  • Server = wrapper + MCP handlers
  • name + description + inputSchema
  • call_tool returns content, not prints
  • Descriptions are how the LLM picks tools

Common Mistakes

  • Printing debug logs to stdout on a stdio server
  • Vague tool descriptions so the LLM picks the wrong tool
  • Crashing the process instead of returning a tool error