MCP & Agent-to-Agent Protocols

Understand the protocols that connect AI agents to tools and to each other: MCP internals, building and securing servers, A2A agent cards and tasks, and how the two fit together, with a real server driven over JSON-RPC.

Start course →

Syllabus

Why Agents Need Protocols

  1. The Integration Problem
  2. MCP, A2A and Function Calling: Who Does What
  3. Hosts, Clients, Servers and Peer Agents
  4. When You Do and Do Not Need Protocols

Inside MCP: Messages, Lifecycle and Primitives

  1. JSON-RPC 2.0 in Five Minutes
  2. The Initialize Handshake and Capabilities
  3. Server Primitives: Tools, Resources, Prompts
  4. Transports: stdio and Streamable HTTP
  5. Client Features: Sampling, Roots, Elicitation, Pagination

Building Good MCP Servers

  1. Your First Server With the Python SDK
  2. Designing Tools the Model Can Use
  3. Errors, Validation and Recoverable Failures
  4. Testing and Inspecting MCP Servers

MCP Clients, Hosts and Remote Servers

  1. How a Host Uses MCP Servers
  2. Remote Servers and Authorisation
  3. Using MCP Tools Inside Your Own Agent

Security for Tool-Using Agents

  1. Trusting Servers: Supply Chain and Tool Poisoning
  2. Prompt Injection Through Tool Results
  3. Least Privilege, Allow-Lists and Human Approval
  4. Sandboxing, Secrets and Audit Logging

Agent2Agent (A2A): Agents Working Together

  1. Why Agent-to-Agent Communication
  2. The Agent Card: Discovery
  3. Tasks, Messages, Parts and Artifacts
  4. Discovery, Routing and Delegation
  5. Streaming, Push Notifications and Security in A2A

MCP and A2A Together

  1. A Reference Architecture
  2. Testing and Observing Multi-Agent Systems
  3. Governance: Registries, Versions and Ownership
  4. Protocols Evolve: Staying Current

Putting It Together

  1. Case Study: A Refund Assistant Across Teams
  2. Revision: Cheat Sheet and Self-Check