Model Context Protocol (MCP) for Developers

Class Duration

7 hours of live training delivered over 1-2 days to accommodate your scheduling needs.

Student Prerequisites

Professional software development experience in TypeScript or Python
Familiarity with REST APIs and JSON

Target Audience

Software engineers and platform engineers who want to extend AI coding assistants with internal tools, private data sources, and organizational APIs using MCP. Relevant for anyone building internal developer tooling on top of Claude, GitHub Copilot, or other MCP-compatible AI systems.

Description

The Model Context Protocol (MCP) is an open standard that defines how AI assistants connect to external tools and data sources. This course teaches developers how to both consume MCP servers (connecting your AI assistant to existing internal and third-party tools) and build them (exposing your organization's APIs and data to AI assistants safely and reliably). We cover the MCP specification, the server and client architecture, transport layers (stdio and HTTP/SSE), tool definitions, resource endpoints, and prompt templates. Labs build working MCP servers in TypeScript and Python against realistic internal tool scenarios.

Learning Outcomes

Explain the MCP specification: servers, clients, tools, resources, and prompts.
Connect an AI assistant (Claude, GitHub Copilot) to an existing MCP server.
Build a working MCP server in TypeScript or Python that exposes tools and resources.
Design safe tool schemas with appropriate input validation and error handling.
Implement both stdio (local) and HTTP/SSE (remote/hosted) transport layers.
Apply authentication, authorization, and rate limiting to MCP server endpoints.
Evaluate organizational patterns for sharing and governing internal MCP servers.

Training Materials

Comprehensive courseware is distributed online at the start of class. All students receive a downloadable MP4 recording of the training.

Software Requirements

Node.js 20+ or Python 3.12+, an MCP-compatible AI client (Claude Desktop, VS Code, Visual Studio, a JetBrains IDE, Zed, Cursor, or Windsurf), and Git.

Training Topics

MCP Fundamentals

Why MCP: the context and tool integration problem
MCP specification overview: servers, clients, hosts
Transport layers: stdio vs. HTTP/SSE
MCP vs. function calling: key differences

Consuming MCP Servers

Connecting MCP-aware hosts: Claude Desktop, VS Code, Visual Studio, JetBrains IDEs, Zed, Cursor, and Windsurf
Browsing and invoking tools from an AI client
Reading resources and using prompt templates
Debugging connection and tool invocation issues

Building MCP Servers in TypeScript

Project setup with the MCP TypeScript SDK
Defining tools: name, description, input schema
Returning structured results and errors
Exposing resources with URI templates

Building MCP Servers in Python

Project setup with the MCP Python SDK
Tool decorators and function handlers
Async patterns for I/O-bound tools
Resource and prompt template definitions

Authentication and Security

MCP server authentication patterns
OAuth 2.0 integration for hosted servers
Input sanitization and safe execution
Preventing prompt injection through tool outputs

Transport: Hosting and Distribution

Local stdio servers: installation and config
Remote HTTP/SSE servers: deployment and TLS
Packaging and distributing MCP servers to a team
Registering servers in Claude Desktop, Copilot (VS Code/Visual Studio/JetBrains), Zed, Cursor, and Windsurf

MCP Registry: Public and Private

The public MCP Registry (preview Sep 2025): canonical catalog and API for server discovery
Public vs. private sub-registries for organizational isolation
Server signing, verification, and trust scoring (incoming with the Registry GA)
Upcoming spec evolutions: server-as-agent capabilities and registry maturation

Organizational Patterns

Internal MCP registry for shared servers
Governance: approving and versioning MCP servers
Monitoring tool invocations in production

Workshop

Build and deploy a local MCP server (CRUD tool lab)
Connect to a remote MCP server
Q&A session