Focuses on agent infrastructure, APIs, tool security, and MCP integration, not general prompt-writing or chatbot use.
← All courses
// AIINFRA 300 · Semester 3
Agentic AI & the Model Context Protocol (MCP)
Building, Securing, and Orchestrating Production AI Agents with MCP
This course teaches learners to design, build, and deploy production-grade AI agents using leading provider APIs (OpenAI, Anthropic Claude, AWS Bedrock, Google Vertex, OpenRouter), code-first agent frameworks (LangGraph, CrewAI, OpenAI Agents SDK, Pydantic AI), custom API-backed command-line tools, and the Model Context Protocol (MCP) for standardized tool and context integration. Learners implement agent-native CLIs with structured output, configuration checks, API authentication, and compound commands before exposing the same capabilities through MCP, then add multi-agent orchestration, human-in-the-loop workflows, agent evaluation, observability, and guardrails. The course culminates in a capstone deploying a secure, observable, multi-agent system with both CLI and MCP-facing tool surfaces.
download / canvas package
Canvas IMSCC Export
Canvas IMSCC Export
Download the complete Canvas import package for AIINFRA 300.
01 / outcomes
Outcomes
Course objectives
- Build agents that call multiple LLM provider APIs (OpenAI, Claude, Bedrock, Vertex, OpenRouter) with cost, latency, and observability instrumentation
- Implement core agent patterns including tool calling, ReAct, planning, and memory/state management from first principles
- Construct production agents using code-first frameworks such as LangGraph, CrewAI, OpenAI Agents SDK, and Pydantic AI
- Build API-backed, agent-native CLI tools and expose selected capabilities through MCP servers with proper authentication and security controls
- Orchestrate multi-agent systems with handoffs, human-in-the-loop checkpoints, and evaluate agent performance using observability tooling
Student learning outcomes
- Build and instrument agents that call OpenAI, Claude, Bedrock, Vertex, and OpenRouter for cost, latency, and reliability.
- Implement tool calling, ReAct, planning, and memory/state management from first principles.
- Construct production agents with LangGraph, CrewAI, the OpenAI Agents SDK, and Pydantic AI.
- Build API-backed, agent-native CLI tools and expose selected capabilities through secure MCP servers.
- Orchestrate multi-agent systems with handoffs and human-in-the-loop checkpoints, evaluated with observability tooling.
02 / schedule
16-week schedule
Wk 01
Agentic AI Foundations and the Provider Landscape
Introduces agentic AI concepts and surveys the provider landscape (OpenAI, Claude, Bedrock, Vertex, OpenRouter) that agents will call.
Wk 02
Tool Calling, ReAct, and Agent Planning Patterns
Covers core agent patterns including tool calling, the ReAct loop, and planning strategies from first principles.
Wk 03
Memory and State Management for Agents
Covers memory and state management techniques that let agents retain context across multi-step interactions.
Wk 04
OpenAI, Claude, and Multi-Provider API Integration
Covers integrating OpenAI and Anthropic Claude APIs into agents alongside multi-provider integration patterns.
Wk 05
AWS Bedrock, Google Vertex AI, and OpenRouter Routing
Covers integrating AWS Bedrock and Google Vertex AI and routing requests through OpenRouter.
Wk 06
Cost, Latency, and Observability Instrumentation
Covers instrumenting agent calls for cost, latency, and observability across providers.
Wk 07
Code-First Agent Frameworks I: LangGraph and Pydantic AI
Introduces building production agents with the code-first LangGraph and Pydantic AI frameworks.
Wk 08
Code-First Agent Frameworks II: CrewAI and OpenAI Agents SDK
Midterm week: covers CrewAI and the OpenAI Agents SDK alongside the course midterm assessment.
Midterm · covers Wks 1–7Wk 09
MCP Architecture Deep Dive
Provides a deep dive into Model Context Protocol architecture for standardized tool and context integration.
Wk 10
Building Agent-Native CLI Tools and MCP Servers
Students map a real API into a reliable CLI with doctor checks, structured output, and compound commands, then expose selected capabilities through MCP.
Wk 11
MCP Authentication and Tool/Agent Security
Covers MCP authentication mechanisms and security controls for tools and agents.
Wk 12
Multi-Agent Orchestration and Handoffs
Covers orchestrating multi-agent systems, including handoffs between cooperating agents.
Wk 13
Human-in-the-Loop Workflows and Emerging Standards (A2A, AGENTS.md)
Covers human-in-the-loop checkpoints alongside emerging agent standards such as A2A and AGENTS.md.
Wk 14
Agent Evaluation and Observability
Covers evaluating agent performance and instrumenting agents with observability tooling.
Wk 15
Agent Deployment and Guardrails Introduction
Covers deploying agents to production and introduces guardrails for safe agent behavior.
Wk 16
Capstone Project & Course Review
Final capstone week: students deploy a secure, observable, multi-agent MCP-based system and review the course.
Capstone03 / Senate-ready controls
Course evidence and boundaries
These controls make the curriculum reviewable without changing the 10-course sequence.
API-backed CLI or service with MCP-facing tool surface, structured outputs, and observability traces.
Tool allowlist, authentication, human approval boundary, secrets handling, and misuse test notes.
Repository, MCP/server documentation, demo transcript, security review, traces/logs, and runbook.
04 / tools
Tools & frameworks
Provider APIs
Agent Frameworks
CLI and MCP Tooling
Languages/Runtimes
Observability
Cost/Latency Monitoring
Orchestration/Deployment
Security
Dev Environment