Understanding MCP

Repository for Understanding MCP (Model Context Protocol) - A hands-on guide

These instructions will guide you through configuring a GitHub Codespaces environment that you can use to run the code.

Prefer to run the labs on your own machine instead? See LOCAL_SETUP.md for the full list of prerequisites and setup steps. The Codespace is the environment the labs are verified against, so a few things will look different locally.

Updated for MCP specification revision 2026-07-28. This is a breaking protocol change: the initialize handshake and Mcp-Session-Id are gone, servers can no longer push requests to clients, and Roots, Sampling and Logging are deprecated. See the "Parts of 2026-07-28 we did not lab" appendix at the end of labs.md, and MCP-2026-07-28-UPDATE-REPORT.md for the full rundown.

The labs run on FastMCP 4, pinned exactly in requirements.txt (currently 4.0.3, the stable release line - FastMCP 4.0 went stable on 2026-08-31). FastMCP 3.4.x hard-pins mcp<2.0 and cannot speak this protocol revision.

1. Set a longer timeout for the environment we'll be using - GitHub Codespaces.

When logged into GitHub, go to https://github.com/settings/codespaces.

Scroll down and find the Default idle timeout section and set it to at least 60 minutes.

2. Click on the button below to start a new codespace from this repository.

Click here ➡️

3. Then click on the option to create a new codespace.

This will run for a few minutes while it gets the virtual environment ready. When the main portion is done, it will still need to do some post-processing. That will look like this:

It will be done when you see output similar to this in the terminal.

4. Run the warmup script in the TERMINAL to make the LLM interactions faster.

python scripts/warmup.py &

5. Open the labs file.

You can open the labs.md file either in your codespace or in a separate browswer tab/instance.**