A plain fetch hands the model the whole document — chrome, scripts, cookie banners — and most of the tokens go to noise. Readability is the same extraction Firefox uses for reader mode, so what comes back is the text a person would actually read, with a byline and site name attached for citation.
An MCP server that runs Mozilla's Readability algorithm over a URL and converts what survives into Markdown, returning the article body plus its metadata instead of raw HTML.
- `parse` — takes a `url` and returns the article `title`, the `content` as Markdown, and metadata: `excerpt`, `byline` and `siteName`
- Ads, popups, navigation menus and footers are stripped by Readability before conversion, which is also what keeps the token count down
- Consistent Markdown formatting rather than whatever the page's HTML happened to be
An MCP client. Install with `npm install server-moz-readability`, or add `npx -y server-moz-readability` to `claude_desktop_config.json`; Smithery can do it for Claude Desktop in one command. No account, no key. It builds on @mozilla/readability, turndown, jsdom and axios. Package version 1.0.3. MIT.
One command — npx -y server-moz-readability