Extra prerequisites

Do this after Exercise 0 · Prerequisites, at home, before the workshop.

Goal: install Docling, download its models and bring in the material for the mod lessons, so everything is ready on the day.

Plan for about 30 minutes and about 1.5 GB of downloads.

Caution

Don’t leave this for the day of the workshop. Docling downloads ~800 MB of Python packages and models. With the whole room sharing the venue wifi there is no time for it, and without Docling you can’t do The Engineering Reviewer with Docling MCP.

Pick one of the two routes. Both start with the repo you created from the template in Exercise 0, and both end with the same check.

Route	What it is	Recommended?
Local, without a container	VS Code on your own computer, with the repo cloned	Yes: faster, and nothing to lose if a codespace stops
Codespaces	The codespace from Exercise 0, as the base workshop does	Works too

Note

These steps have been tested on Windows 11. The macOS (Apple Silicon) steps have not been verified end to end yet. If something doesn’t match on your Mac, open an issue in the workshop repo.

Copilot plan and model

The workshop needs the same Copilot plans as the base workshop: Student, Pro, Pro+, Business or Enterprise. Copilot Free is not supported. If you are a student, apply for the GitHub Student Developer Pack, which includes Copilot Student. Verification can take a few days, so do it as soon as possible.

Note

Always pick the Auto model in the chat, even when a base exercise asks for another one. Exercise 2 asks for Claude Sonnet 4.5, which has been retired. The whole workshop has been tested with Auto, and it is the only option with Copilot Student.

Route A: local, without a container

Every command runs in a terminal: Windows Terminal (PowerShell) on Windows or Terminal on macOS. After installing each tool, open a new terminal so that its PATH is up to date.

1. VS Code

Install or update VS Code to version 1.112 or later. Check it in Help > About (on macOS, Code > About Visual Studio Code). Copilot Chat is built in: open the Chat view from the Copilot icon in the title bar and sign in with your GitHub account.

2. Git and Node.js

Tailspin Toys needs Node.js 22.13 or later, because it uses the node:sqlite module.

1. Check what you have:

   git --version
   node --version

2. If git is missing, install it from git-scm.com. On macOS you can also run xcode-select --install.

3. If node is missing or older than 22.13, download the LTS version from the Node.js download page and run the installer with the default options. On Windows, keep Add to PATH checked.

4. Open a new terminal and run node --version again.

3. GitHub CLI (gh)

Matt Pocock’s workflow uses gh to read and label the issues of your repo.

1. Install it:

   # Windows
   winget install --id GitHub.cli

   # macOS
   brew install gh

   If you don’t use winget or Homebrew, download the installer from cli.github.com.

2. Open a new terminal and sign in. Accept the defaults (GitHub.com, HTTPS, authenticate Git with your GitHub credentials and log in with a web browser):

   gh auth login

4. Clone your repo

Clone the repo you created in Exercise 0 into a short path and open it in VS Code. Replace YOUR_USER/YOUR_REPO with the repository path you noted down in Exercise 0:

\dev

mkdir C:\dev -Force
cd C:\dev
gh repo clone YOUR_USER/YOUR_REPO
cd YOUR_REPO
code .

# macOS
mkdir -p ~/dev && cd ~/dev
gh repo clone YOUR_USER/YOUR_REPO
cd YOUR_REPO
code .

When VS Code asks whether you trust the authors of the folder, answer Yes, I trust the authors. In Restricted Mode the MCP servers of .vscode/mcp.json don’t start.

5. App dependencies and the Playwright browser

Open the integrated terminal of VS Code (Terminal > New Terminal) and run:

npm install
npx playwright install chromium

The second command downloads the Chromium (~300 MB) that Copilot uses in Exercise 3 · Playwright MCP.

You can check that the app works with npm run dev: it shows the Tailspin Toys catalog at http://localhost:4321. Stop it with Ctrl+C.

6. Run setup

Go on to Set up the mod.

Route B: Codespaces

1. Do Exercise 0 as it is: your codespace already has Node, git and gh.
2. Open the terminal of your codespace (Terminal > New Terminal) and go on to Set up the mod.

Caution

Use this same codespace on the day of the workshop, and don’t rebuild it. Docling and its models are installed in your home folder, outside /workspaces, so a new or rebuilt codespace loses them. A codespace that stops after a while of inactivity keeps them: just open it again from the Code > Codespaces tab of your repo.

Set up the mod

The mod has a small command-line tool that does everything in one go. From the root of your repo, in the VS Code terminal, run:

npx -y github:emepetres/copilot-devdays-workshop#mod-v1 setup

It works the same in PowerShell, bash and zsh. setup:

1. Copies the material of every mod lesson (skills, documents, diagrams…) to its place in your repo. Leave it uncommitted: each lesson tells you when to commit it. It never overwrites a file you already have.
2. Installs uv with its official installer, if you don’t have it. uv installs Python tools; you don’t need Python.
3. Installs Docling MCP 3.3.0, the exact version of the workshop. On Linux and Codespaces it installs the CPU build of PyTorch, to avoid ~2.8 GB of GPU libraries.
4. Runs a warm-up conversion: it starts docling-mcp-server with the same configuration Copilot will use, converts a two-page PDF and checks that OCR reads a control word on its scanned page. This downloads Docling’s models.

Step	Download	Approximate time
uv	~20 MB	seconds
Docling MCP (Python, PyTorch for CPU and dependencies)	~300 MB (~1 GB installed)	1–5 min
Docling models, during the warm-up	~510 MB	1–10 min

The warm-up shows no progress while it downloads: let it finish. When it is done you will see:

Docling warm-up
✔ docling-mcp-server started (1 s)
⏳ Converting warmup.pdf. The first time it downloads ~510 MB of models and can take several minutes...
✔ PDF converted (74 s)
✔ OCR of the scanned page: "Giralda" found

✅ Setup finished. Restart VS Code (close every window) so it picks up the new PATH,
then run check from its terminal.

Restart VS Code, closing every window, so that it finds docling-mcp-server on its PATH. In Codespaces, reload the window instead: open the Command Palette (Ctrl+Shift+P) and run Developer: Reload Window.

Tip

If Windows PowerShell refuses to run npx because running scripts is disabled on this system, use npx.cmd with the same arguments.

Enable agent plugins

Matt Pocock’s workflow installs Matt Pocock’s skills as an agent plugin, which needs a setting:

1. Open the settings with Ctrl+, (on macOS, Cmd+,).
2. Search for chat.plugins.enabled and check the box.

Final check

Open a new terminal in VS Code, at the root of your repo, and run:

npx -y github:emepetres/copilot-devdays-workshop#mod-v1 check

You are ready when every line is green:

✅ Node 22.20.0
✅ uv 0.9.24
✅ docling-mcp 3.3.0
✅ Docling models
✅ Lesson material

docling-mcp-server: C:\Users\you\.local\bin\docling-mcp-server.exe

All set for the workshop.

Each ❌ says what to run to fix it. Keep the docling-mcp-server path at hand: if VS Code can’t find the server in the Docling lesson, you will use that path. If something goes wrong on the day, check is the first thing we will ask you to run.

If something fails

• uv is installed but the terminal can’t find it, or the warm-up says docling-mcp-server is not installed: close every terminal (also the one in VS Code), open a new one and run setup again. It doesn’t repeat what is already done.

• The download stops halfway: run the warm-up again. The models already downloaded are not downloaded again:

  npx -y github:emepetres/copilot-devdays-workshop#mod-v1 warmup

• WinError 206, FileNotFoundError or very long paths on Windows: see Long paths on Windows.

• OCR doesn’t find the control word: copy the error and the log lines the command prints, and open an issue in the workshop repo.

• Mac with an Intel processor: PyTorch no longer publishes builds for Intel Macs, so Docling may not install. If that happens, tell us at the workshop: you will do the Docling lesson in pairs with someone who has it working.

• npx github: doesn’t work at all: you can still copy the material with npx -y degit@3.10.1 --force emepetres/copilot-devdays-workshop/mod/lessons/<lesson>#mod-v1 . for each lesson (pocock, images, docling, extra), and install Docling by hand with uv tool install "docling-mcp[local]==3.3.0" (on Linux and Codespaces, add --torch-backend cpu).

Long paths on Windows

Windows limits paths to 260 characters by default, and some Python packages and Docling models have very deep paths. To avoid problems:

• Clone the repo into a short path, such as C:\dev\tailspin-toys, not into a deeply nested folder.

• Leave uv and Hugging Face in their default folders: don’t change UV_CACHE_DIR, HF_HOME or HF_HUB_CACHE.

• If you still see WinError 206 or FileNotFoundError with very long paths, enable long paths. Open PowerShell as administrator, run these two commands, restart your computer and run setup again:

  New-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1 -PropertyType DWORD -Force
  git config --global core.longpaths true

Summary

You are ready for the workshop. You have:

• the tools of your route: VS Code, Node.js, git and gh, or your codespace;
• the material of the mod lessons in your repo, uncommitted;
• Docling installed and its models downloaded;
• agent plugins enabled and the Auto model in mind;
• check all green.

On the day of the workshop you will start with Exercise 1 · Custom instructions.

Resources

• GitHub Copilot plans
• Node.js download
• Installing uv
• Docling MCP