Agentic AI with FastMCP and A2A is a hands-on course in using Python to build MCP servers and clients with FastMCP, then using the A2A standard so several agents work together as a multi-agent system. It suits developers, students and automation practitioners with basic Python, and you leave able to deploy an MCP server to the cloud.
AI today is more than a chatbot that answers questions. Agentic AI can plan and carry out work for us, built around two open standards: MCP (Model Context Protocol), which gives AI a single standard way to use tools, files and databases, and A2A (Agent2Agent), which lets agents from different systems discover each other and hand work over. Both standards move quickly, so many examples from a year ago still use SSE or older APIs that have since been retired.
This course shows learners how to build an agentic AI system in Python from scratch with FastMCP 3: creating MCP servers with tools, resources and prompts, connecting files and databases, using them from Claude Desktop, serving them over Streamable HTTP and building an MCP client that works with Gemini or OpenAI. Learners then build agents to the A2A 1.0 standard with the Python SDK, make several agents work together safely and deploy an MCP server to the cloud. The course closes with a multi-agent trip planner workshop. (2 days, 6 hours per day, 12 hours in total, Intermediate level.)
Course Details
This course is the 2026 edition of Building Agentic AI with FastMCP and A2A. It has been rewritten for FastMCP 3, Streamable HTTP in place of SSE, the 2026-07-28 MCP specification and A2A 1.0. Deployment now uses Prefect Horizon, and a new topic covers agent security. It runs for 2 days, 6 hours per day (12 hours in total, 09:00-16:00), as lectures with Python labs in every topic. Intermediate level. Learners use their own Gemini or OpenAI API key and take home the code for every lab.
Section 1: Foundations of Agentic AI, MCP and A2A
Section 2: Lab: A First MCP Server with FastMCP 3
Section 3: Lab: Resources, Prompts and Context
Section 4: Lab: Connecting Databases Safely
Section 5: Lab: Streamable HTTP and Authentication
Section 6: Lab: An MCP Client in Python
Section 7: Meet the A2A Protocol 1.0
Section 8: Lab: A First A2A Agent with the Python SDK
Section 9: Lab: An A2A Agent That Uses MCP Tools
Section 10: Designing Multi-Agent Systems Safely
Section 11: Lab: Deploying an MCP Server to the Cloud
Section 12: Workshop: Multi-Agent Trip Planner
No public rounds are open right now. Join the waiting list and we will contact you first when the next round opens, or ask us on LINE. Or call 02-570-8449 or 088-807-9770
MCP, the Model Context Protocol, is a standard that lets AI applications use tools, files and databases through servers in one consistent way, rather like plugging devices into the AI. A2A, Agent2Agent, is a standard that lets agents built on different frameworks or by different organisations find each other through Agent Cards, then hand over work and receive results as tasks and artifacts. Real systems often use both: agents talk to each other over A2A, and each agent uses its tools through MCP.
Install FastMCP with uv or pip, create a FastMCP object with a server name, then write ordinary Python functions decorated with @mcp.tool. FastMCP turns type hints and docstrings into a schema the AI understands. Add @mcp.resource for data the AI can read and @mcp.prompt for ready-made prompts, then call mcp.run() to run over STDIO, or pass transport="http" to serve it on the web. You can test it with MCP Inspector or connect it to Claude Desktop.
With STDIO, the AI application runs the server as a program on the same machine and talks to it through standard input and output, which suits personal tools such as Claude Desktop. Streamable HTTP serves the server as a web service that many clients can reach over the network, which suits cloud deployment. SSE is the older web transport that Streamable HTTP replaced, and FastMCP marks it as deprecated, so use it only when connecting to legacy systems.
It suits software and AI developers who want to build agentic AI that works in practice, automation practitioners who want AI to use their organisation's systems, AI students, and tech leads setting standards for connecting AI to tools. You should write basic Python, be able to use a terminal and VS Code, and have your own Gemini or OpenAI API key. No web development background is needed.
You will be able to build MCP servers with tools, resources and prompts using FastMCP 3, connect files and databases safely, serve over Streamable HTTP, build an MCP client that lets Gemini or OpenAI call tools, build A2A agents with the Python SDK, design a multi-agent system with an orchestrator and deploy an MCP server to the cloud. Participants who complete the training receive a Certificate of Completion with its own number, which can be verified online.
Bring a Windows, macOS or Linux laptop with permission to install software, able to run Python, uv and VS Code. You need your own Google Gemini or OpenAI account and API key, which may incur usage charges, and a GitHub account is useful for deploying to Prefect Horizon on its free personal tier. Claude Desktop and MCP Inspector are installed together in class.
Web design today has to look good, work on every screen size, load quickly and be easy for a team to keep editing. Writing all the CSS by hand often ends in long files nobody dares to touch, while component-based CSS frameworks make sites look alike and are…
A good dashboard does more than show what sales were. It should help answer why the numbers changed, which customers are about to leave and which points are unusual enough to need attention now. Power BI has built-in AI visuals that answer these questions…
Angular is the framework many organisations choose for medium to large web applications, thanks to its clear structure and complete built-in tooling. Angular has changed a great deal in recent years, with standalone components replacing NgModules, signals,…
