A self-paced course on building AI agent systems with MCP (Model Context Protocol) and A2A (Agent-to-Agent). Learn how LLMs connect to tools and how agents delegate to each other.
| Module | What You'll Learn | Quick Start |
|---|---|---|
| MCP | Tools, resources, prompts — LLMs calling external capabilities | cd MCP && pip install -r requirements.txt && python examples/01-calculator/server.py |
| A2A | Agent orchestration — agents delegating to specialized sub-agents | cd A2A && pip install -r requirements.txt — see A2A Quick Start |
Goal: Expose tools (calculators, file readers, databases) to LLMs via a standard protocol.
- Build MCP servers with FastMCP
- Understand tools vs resources vs prompts
- Connect to Cursor, VS Code, or Claude Desktop
→ MCP README — full learning path, examples, exercises
Goal: Build agents that call other agents over HTTP. One coordinator delegates to specialized remote agents (math, prime checker, etc.).
- Use Google ADK with Azure OpenAI or Gemini
- Expose agents with
to_a2a(), consume withRemoteA2aAgent - Static vs dynamic orchestration (registry-based)
→ A2A README — architecture, quick start, teaching docs
ai-agent-systems-course/
├── MCP/ # Model Context Protocol — tools for LLMs
│ ├── docs/ # Teaching docs
│ ├── examples/ # Calculator, file reader, SQLite, etc.
│ ├── exercises/ # Hands-on tasks
│ └── solutions/ # Exercise answers
├── A2A/ # Agent-to-Agent — agent orchestration
│ ├── docs/ # Teaching docs (9 sections)
│ ├── remote_math_agent.py
│ ├── remote_prime_agent.py
│ ├── coordinator_agent.py
│ └── dynamic_coordinator_agent.py
└── README.md
- Python 3.10+
- MCP: Basic Python (no prior MCP knowledge)
- A2A: API keys for Azure OpenAI or Google Gemini — see A2A Environment Setup
See MCP/LICENSE and A2A materials for licensing terms.