TermShelf Claude Code plugin

Two skills for working with TermShelf, a Legal Content Operations system:

Skill	Audience	What it does
termshelf	Developers	Generate read-only integration code (Next.js / Astro / Express / Laravel / …) that consumes already-published legal texts via the TermShelf Public Delivery API. No credentials needed.
termshelf-author	Operators	Drive a brand-onboarding or document-drafting workflow end-to-end against the TermShelf Management API — create brands, sites, domains, document types, full documents (sections + blocks with snippet references), and per-locale variable / snippet overrides via an MCP server. Requires a token. Stops before publishing.

Install

In Claude Code:

/plugin marketplace add termshelf/claude-plugin
/plugin install termshelf@termshelf

Or for local testing — from inside Claude Code, point /plugin marketplace add at a local clone:

/plugin marketplace add /absolute/path/to/termshelf-claude-plugin
/plugin install termshelf@termshelf

Tier 1 — termshelf (developer-facing)

Ask Claude in plain language:

Add my TermShelf privacy policy to this Next.js app.

Embed the TermShelf imprint as an HTML fragment into my Astro site,
caching for 5 minutes.

Generate a Laravel Blade partial that fetches my TermShelf terms
in JSON and renders them with my own template.

Claude will ask for whichever of these it still needs (account hash, site slug, document type code, target locale, …) and produces a working integration with proper ETag caching, framework-idiomatic placement, and a starter stylesheet for the ts-document classes.

Shortcut: open your TermShelf customer-app → avatar dropdown → "Integration reference". The page lists everything Claude needs on one screen. Click "Copy integration context" and paste the blob into the chat — Claude parses it and skips the questions.

This tier does NOT call any TermShelf service on your behalf. It generates code locally.

Tier 2 — termshelf-author (operator-facing)

This tier does call TermShelf — it drives the /api/v1/management/* surface programmatically.

One-time setup

After installing the plugin, install the MCP server's Node dependencies:

cd "$(claude code plugins root)/termshelf/mcp-server"   # or your manual clone path
npm install

Then issue a Management API token in the customer-app (Settings → API Tokens) with the scopes you want Claude to have:

Scope	What it grants
structure:read	List sites + their domains, markets, profiles
content:read	List workspace variables, snippets, locales, documents, document types
structure:write	Create brands, sites, domains, attach markets
overrides:write	Create per-locale variable and snippet overrides
content:write	Create / edit workspace snippets + variables, document types, full documents (sections + blocks)

For the typical brand-onboarding workflow, the first four. For drafting whole documents, add content:write. Do NOT include publish:trigger — publishing stays in your hands inside the customer-app.

Export the token before launching Claude Code:

export TERMSHELF_TOKEN=<your-token>
# For local development:
# export TERMSHELF_BASE_URL=http://app.termshelf.local:9191

The MCP server reads the token from the environment and never echoes it. If you ever see it in a Claude transcript, rotate it.

Use

Two flavours of prompt are supported:

Onboard "Acme Legal GmbH" (https://acme.de) as a new brand under my workspace.
Create the site, domain, and the per-locale variable + snippet overrides
needed for the documents we already have. Stop before publishing.

Draft a Datenschutzerklärung document of type "privacy" for the Acme brand.
Use our existing gdpr_rights_clause snippet for the "Rechte der Betroffenen"
section and pull company_name + contact_email from variables. Stop before
publishing.

Claude will:

1. Discover existing variables / snippets / locales / documents / document types in your workspace.
2. Fetch the brand's website and (if needed) ask you for missing legal entity details, OR analyse the requested document's structure.
3. Produce a structured plan as a markdown table.
4. Wait for your explicit approval.
5. Execute the plan — brand, site, domain, market, document, sections, blocks, variable + snippet overrides.
6. Verify each (document, brand, locale) against the live preview and iterate.
7. Hand back to you with a link to review the result and hit Publish in the customer-app.

See skills/termshelf-author/SKILL.md for the full workflow + error-handling contract.

What's inside

.
├── .claude-plugin/
│   ├── marketplace.json        # marketplace catalog (single plugin)
│   └── plugin.json             # plugin manifest + MCP server registration
├── mcp-server/                 # Tier 2 — Node MCP server
│   ├── index.mjs               # tool definitions for /api/v1/management/*
│   ├── package.json
│   └── README.md               # setup + tool reference
└── skills/
    ├── termshelf/              # Tier 1 — read-only integration helper
    │   ├── SKILL.md
    │   ├── reference/
    │   │   ├── api-contract.md
    │   │   ├── caching.md
    │   │   └── styling.md
    │   └── templates/
    │       ├── nextjs.tsx
    │       ├── astro.astro
    │       ├── express.js
    │       ├── laravel.blade.php
    │       └── styles.css
    └── termshelf-author/       # Tier 2 — operator-facing workflow
        └── SKILL.md

Boundaries

• Tier 1 (termshelf) is read-only on the developer's side. It does not call any TermShelf API and does not require a token.
• Tier 2 (termshelf-author) calls the Management API on your behalf, but only via the MCP server's typed tools. It does NOT call any read-only delivery endpoint. It does NOT publish.
• Tier 2 deliberately leaves publishing to the operator. The token must not carry publish:trigger.
• The two skills coexist; Claude picks the right one based on whether you're trying to embed published content (Tier 1) or author content / structure (Tier 2).

License

MIT — see LICENSE.