data-liberation

data-liberation-agent

Extract content from closed web platforms into WordPress-compatible WXR files.

The problem

Closed platforms make it hard to leave. Wix has no HTML export and caps RSS at 20 posts. JavaScript-rendered content and limited APIs leave your site data locked inside.

The solution

This tool extracts all content from closed platforms — posts, pages, media, navigation, redirects, products — and produces a standard WordPress WXR file ready to import.

Where to host WordPress: If your current provider also offers WordPress, you can move to WordPress and stay with them. WordPress.com is another option: the $4/mo Personal plan now supports plugins and themes, and the WordPress.com MCP integration gives AI agents direct write access.

Supported platforms

Platform	Status	Prompt
GoDaddy Websites & Marketing (pages/blog)	Ready	prompts/godaddy-wm.md
Hostinger Website Builder (blog/pages/products)	Ready	—
HubSpot	Ready	—
Shopify (blog/pages/products)	Ready	prompts/shopify.md
Squarespace	Ready	prompts/squarespace.md
Webflow	Ready	prompts/webflow.md
Weebly (blog/pages/products)	Ready	—
Wix	Ready	prompts/wix.md
Any other website (generic fallback)	Best-effort	—

All eight platforms have MCP adapters with full extraction support including products (exported as WooCommerce-compatible CSV). Sites matching none of them fall back to a generic default adapter that renders each page in a headless browser and extracts the main content, media, and any JSON-LD products — best-effort, since it can't key off platform-specific markup. GoDaddy Websites & Marketing is pages + blog only in v1; GoDaddy Online Store (OLS) product support is planned for v1.1.

Getting started — agent-first

data-liberation-agent is built to be driven by an AI agent. The front door is the liberate skill: one command runs the full pipeline — detect the platform, inventory every page/post/product, extract content and media, capture screenshots and design tokens, then reconstruct the site as an editable WordPress block theme and import it into a local WordPress preview.

Studio required for preview/import. Install Automattic Studio before running /data-liberation:liberate. Extraction itself needs no WordPress.

Claude Code

Install from the marketplace:

claude plugin marketplace add Automattic/data-liberation-agent
claude plugin install data-liberation@data-liberation

Or from a local checkout (for development on the plugin itself):

cd data-liberation-agent
claude plugin marketplace add .
claude plugin install data-liberation@data-liberation

Then, in Claude Code:

/liberate https://your-site.com

What you'll see: the agent detects the platform, inventories all pages/posts/products, pauses to confirm scope and estimated time, then extracts content and media. It then drives the design phase — clustering page layouts, building a responsive block theme that mirrors your source site's structure and visual style, and importing everything into Automattic Studio. When it finishes you get a local preview URL and a run-report.json summarizing what was built, what's faithful, and any gaps.

The result is a responsive, editable WordPress block theme — not a static copy.

Codex

cd data-liberation-agent
codex

The .codex-plugin/plugin.json and .mcp.json register the MCP server and skills automatically. The liberate flow runs sequentially on Codex (the builder fan-out step degrades to a sequential loop).

Then in Codex:

$liberate https://your-site.com

Gemini CLI

cd data-liberation-agent
gemini extension link .

Any MCP client

Run the MCP server over stdio:

npx tsx src/mcp-server.ts

# or

npm run mcp

It exposes 35 tools. The ones you'll call directly for a deterministic extract → QA → import flow:

liberate_detect, liberate_discover, liberate_inspect, liberate_extract, liberate_screenshot, liberate_status, liberate_qa, liberate_verify, liberate_setup, and liberate_import — plus liberate_paths (resolve the output directory) and liberate_probe / liberate_map_apis (browser-based diagnostics). The remaining tools drive the design/reconstruction phase and are orchestrated by the skills rather than called by hand. Full reference with parameters: docs/mcp.md.

Output