What it fetches and sends.
The skill runs in your project and sends nothing of yours. The MCP sees only the calls your agent makes to it.
Updated . Struq skill 0.5.0.
The skill
The skill is instructions, the block files and five small Node scripts (blocks.mjs, canon.mjs, images.mjs, check.mjs, log.mjs) that your agent runs in your project. They read and write files there, including a build log, STRUQ-LOG.md.
- Blocks and the direction ship inside the skill. Nothing is fetched to install a block or write the tokens.
- Fonts load from Google Fonts, or Fontshare for a face that lives there, in the visitor’s browser, as on any site with web fonts.
- Photos, for every slot your own photos do not fill: images.mjs reads the public pack list at https://struq.nl/api/media/packs and downloads the chosen images from https://struq.nl/media/packs/. Offline it writes plates graded to your palette instead.
- The installer downloads the package from npm, looks for your AI tools’ folders in your home folder and copies the skill into them. It sends nothing.
The MCP
If you connect it, your agent calls Struq’s tools with your brief. Nothing else leaves your machine. Struq never reads your code: the check runs on your side.
What Struq stores
| What | Kept for |
|---|---|
| The connection: your client, what you allowed, hashed tokens | Until you revoke it |
| Each tool call: which tool, the outcome, how long it took | One year |
Struq never stores what you ask or what comes back.
Disconnect
Revoke the connection in your Struq account, under AI connections. Access stops at once. The skill needs no account, so there is nothing to disconnect there: remove it with npx @struq/skill@latest remove.