AI agent installation¶

Paste a prompt into an AI coding agent that can run terminal commands. The agent gathers the license, target, project-folder, profile, and consent choices before it runs commands. Review Target, prerequisites, and project folder first for target, prerequisite, and project-folder guidance.

Typical installation

Install all compatible contexts with one setup prompt.

Typical installation prompt

Custom installation

Choose contexts and services interactively.

Custom installation prompt

Typical installation prompt¶

This prompt asks which editor and AI coding agent to use. GitHub Copilot in VS Code is the default, not a requirement. It uses setup to install the contexts and deploy them into a project folder under the current working directory.

Install STM32Cube for Agentic AI for my selected editor and AI coding agent.

Before running any command, ask me these questions in one message and wait
for my answers:
1. Licenses: do I accept the SLA0044 license agreement of the STM32Cube for
   Agentic AI bundles and the license agreements of the tools that they
   install? The commands accept them on my behalf. If STM32CubeCLI must be
   installed, ask whether I accept the SLA0048 license agreement before
   running its installer. Stop if I decline any license.
2. Target: choose one supported combination:
   - GitHub Copilot in VS Code (default)
   - GitHub Copilot CLI: --ide none
   - Claude Code in VS Code: --agentic-system claude-code
   - Claude Code CLI: --ide none --agentic-system claude-code
   - Cursor editor: --ide cursor --agentic-system cursor
   - Cursor CLI: --ide none --agentic-system cursor
   Host extensions required by a selected editor are separate
   prerequisites. The toolkit installer does not install or manage them.
   Do not treat them as context extensions or claim that setup installs
   them.
3. Project folder: what project folder should receive the contexts? Use a
   project or workspace folder, not my home folder.
4. VS Code profile: for a VS Code target, which exact profile should
   receive the extensions? Do not ask this for another target.
5. Consent: do I accept installing bundles and their declared dependencies
   on my machine and deploying files into the selected project folder?
If I do not accept the licenses or the installation, stop.

After my answers, run these steps without asking me anything else. Use the
terminal tool you already have. Stop at the first failure, show me the full
error, and propose a fix.

1. If "cube --version" fails, install STM32CubeCLI with the command for the
   shell of your terminal tool:
   - PowerShell: irm https://developer.st.com/bundles/install/install.ps1 | iex
   - bash or zsh: curl -fsSL https://developer.st.com/bundles/install/install.sh | bash -
     then: export PATH="$HOME/.local/stm32cube/bin:$PATH"
2. From a working directory outside any deployed project's bundle lock,
   run "cube bundle install --yes cube-agentic-ai-installer", then
   "cube bundle update cube-agentic-ai-installer" to use the latest
   available installer even if an older version was already installed.
3. Build the target options from my answer. For any VS Code target, add
   "--install-vscode-extensions". If I chose a named profile, also add
   "--vscode-profile <exact profile name>". Otherwise use the Default
   profile. Run "cube agentic-ai setup --project-folder <project-folder>
   --python venv <target options> <VS Code extension options>". Do not pass
   an installer or context version. setup can take several minutes: allow a
   long timeout or run it in the background and wait. If it stops before
   deploying, run it again. Once it has deployed, run
   "cube agentic-ai doctor <project-folder> --fix" instead.
4. Check that the JSON result has "ok": true and
   "data.status": "ALL_DEPLOYED". ALL_DEPLOYED confirms deployment, not
   dependency readiness. Also inspect "data.readiness",
   "data.contexts[].requirements[]", and "data.nextSteps". If action
   remains, run "cube agentic-ai doctor <project-folder> --fix". For VS Code
   extension installation, add "--vscode-profile <exact existing profile
   name>". Complete manual steps and check again with doctor. Do not rerun
   setup on the deployed project. Show me "data.licenses", "data.notices",
   "data.unavailableContexts", and "data.nextSteps".
5. VS Code only. Setup installs the declared extensions because the command
   includes "--install-vscode-extensions". If "data.vscodeCli.onPath" is
   false, do not run "code" or ask me to install extensions manually. Tell
   me that extension installation could not run because "code" is not on
   PATH. Otherwise, for each entry of "data.vscodeProfiles", run only its
   "verifyCommand". If the profile is not found, stop and ask me to choose
   an existing profile. Inspect "data.vscodeInstall" and "data.nextSteps"
   for the installation result. Do not run "installCommands" yourself.
6. Apply "data.hostInstructionUpdate" as "data.nextSteps" says. For VS Code,
   run "data.openCommand" if present and tell me to continue in the new
   VS Code window.
7. Open the deployed project with the selected editor or start the selected
   CLI agent from that project folder. Approve its project MCP servers as
   required by that agent. Ask it to list the STM32 services deployed in
   this project, then make one read-only request using a deployed context.
   Ask it to cite its source or evidence. Do not change files or operate a
   device during this first-use check.

Never ask me for a password or a token in the chat.

setup can target GitHub Copilot, Claude Code or Cursor, in an editor or in a terminal (see Target, prerequisites, and project folder), a project folder not yet deployed by this installer, or a selection of contexts: answer question 2 with your choice, or add your choices to the prompt, for example folder C:\work\demo, only the Embedded Software and STM32CubeMX2 contexts. The agent adds the matching options, which the customization section lists. For an already deployed project, use project-add or project-remove followed by doctor --fix instead of setup.

For copyable setup commands for all six targets, see Setup commands in the manual installation.

Expected interaction¶

The agent asks for licenses, target, project folder, VS Code profile when
relevant, and consent before it runs a command. For VS Code targets, setup installs declared extensions in the selected existing profile.
The AI coding agent can request an approval for each terminal command. These approvals depend on the permission settings of the agent, not on the installer.
setup writes progress lines while it installs, and prints one JSON result at the end.
After setup, continue in the selected editor or start the selected CLI agent
from the project folder. Approve project MCP servers as required by that agent, then verify the deployed services with a read-only request. The exact approval steps depend on the target. See the supported target details.

For a VS Code target, keep the project folder as the top-level workspace folder. See Project and workspace folder.

Custom installation prompt¶

Use this prompt to see the installation in two steps and to choose the contexts and services one by one. The first step installs all contexts on the machine (machine-bootstrap). The second step deploys the ones you choose into a project (project-deploy-installed) and asks about each context and each service.

Install STM32Cube for Agentic AI step by step, with my choice of contexts
and services.

Before running any command, ask me these questions in one message and wait
for my answers:
1. Licenses: do I accept the SLA0044 license agreement of the STM32Cube for
   Agentic AI bundles and the license agreements of the tools that they
   install? The commands accept them on my behalf. If STM32CubeCLI must be
   installed, ask whether I accept the SLA0048 license agreement before
   running its installer. Stop if I decline any license.
2. Target: choose one supported combination:
   - GitHub Copilot in VS Code (default)
   - GitHub Copilot CLI: --ide none
   - Claude Code in VS Code: --agentic-system claude-code
   - Claude Code CLI: --ide none --agentic-system claude-code
   - Cursor editor: --ide cursor --agentic-system cursor
   - Cursor CLI: --ide none --agentic-system cursor
   Host extensions required by a selected editor are separate
   prerequisites. The toolkit installer does not install or manage them.
   Do not treat them as context extensions or claim that setup installs
   them.
3. VS Code profile: for the VS Code target, do I use the Default profile or
   another profile? If another, what is its exact name? Do not ask this for
   another IDE target.
4. Project folder: what project folder should receive the contexts? Use a
   project or workspace folder, not my home folder.
5. Consent: do I accept that you install software on my machine, in my user
   account and without administrator rights, and create the project folder
   if it does not exist?
If I do not accept the licenses or the installation, stop.

After my answers, run these steps. Use the terminal tool you already have.
Stop at the first failure, show me the full error, and propose a fix.

1. If "cube --version" fails, install STM32CubeCLI with the command for the
   shell of your terminal tool:
   - PowerShell: irm https://developer.st.com/bundles/install/install.ps1 | iex
   - bash or zsh: curl -fsSL https://developer.st.com/bundles/install/install.sh | bash -
     then: export PATH="$HOME/.local/stm32cube/bin:$PATH"
2. From a working directory outside any deployed project's bundle lock,
   run "cube bundle install --yes cube-agentic-ai-installer", then
   "cube bundle update cube-agentic-ai-installer" to use the latest
   available installer even if an older version was already installed.
3. Run "cube agentic-ai machine-bootstrap --upgrade-installed-in-range".
   It installs all contexts and their tools on the machine, and nothing in
   any project.
4. Create the project folder if it does not exist: "mkdir -p" in bash or
   zsh, "New-Item -ItemType Directory" in PowerShell. If it exists and is not
   the project I named, ask me before using it.
5. Run "cube agentic-ai project-deploy-installed <folder>
   --mode interactive --services-mode interactive <target options>". Target
   options: none for GitHub Copilot in VS Code, "--ide none" for GitHub
   Copilot CLI or app, "--agentic-system claude-code" for Claude Code in
   VS Code, "--ide none --agentic-system claude-code" for Claude Code CLI,
   "--ide cursor --agentic-system cursor" for the Cursor editor,
   "--ide none --agentic-system cursor" for Cursor CLI, "--canonical" for
   another agentic system. The installer asks, for
   each context and each service, whether to deploy it. Relay each question
   to me, and type my answer in the terminal. Do not answer for me. Required
   items are deployed without a question.
6. Check that the results of steps 3 and 5 have "ok": true and
   "data.status": "ALL_DEPLOYED". Run "cube agentic-ai doctor <folder> --fix":
   the deployment does not install what the contexts need, such as Python
   packages or the STM32CubeMX2 application. For every VS Code target,
   including Claude Code and GitHub Copilot, add
   "--vscode-profile <exact existing profile name>". For the Default
   profile, explicitly use '--vscode-profile "Default"'. Without this
   option, doctor does not install declared extensions. Show me
   "data.readiness", "data.nextSteps", and "data.vscodeInstall" when present.
7. VS Code only. If "data.vscodeProfile" is absent, skip the extension
   commands. Otherwise, if "code --version" fails, stop and tell me extension
   installation could not run because "code" is not on PATH. Do not ask me
   to install extensions manually. Check "data.vscodeInstall": claim
   success only when its "status" is "PROCESSED" and none of its
   "extensions[]" entries has "outcome": "FAILED". If the profile was not
   found, stop and ask me to choose an existing profile. For another repair
   failure or a missing result, stop, show the reported details and
   "data.nextSteps", and propose a fix. After successful repair, run
   "data.vscodeProfile.verifyCommand". Do not repeat its "installCommands".
   Run "data.vscodeProfile.openCommand" only after the profile is verified.
   Extension installation does not create ".vscode/extensions.json" in
   this custom route. Do not use that file's presence or absence as proof
   that extensions are installed. For another target, tell me how to open
   the folder with it.
8. Use the selected agent with the project folder open. Approve its project
   MCP servers as required by that agent. Ask it to list the STM32 services
   deployed in this project, then make one read-only request using a
   deployed context. Ask it to cite its source or evidence. Do not change
   files or operate a device during this first-use check.

Never ask me for a password or a token in the chat.

For the same choices in a terminal, see Custom installation in the manual installation. To update an existing installation, see the installer user manual. For target-specific MCP approval steps, see the supported target details.

Troubleshooting¶

Agent asks for a password or a token

Type credentials directly in the terminal, never in the chat. Then tell the agent to continue.

A step fails or times out

The agent stops and shows the error. After the fix, ask the agent to run the failed step again. machine-bootstrap and doctor --fix can run again safely. setup runs only in a project folder that has never been deployed: once it has deployed, finish with doctor <project folder> --fix.

The agent says that code is not on the PATH

The VS Code extensions are not necessarily missing: the agent could not check them. Add code to the PATH, or install the extensions from the Extensions view of VS Code. See the VS Code command line reference.

For other issues, see the troubleshooting section of the manual installation.