#!/usr/bin/env node /** * Convert MDX files to clean markdown for LLM consumption * Removes JSX components and converts to pure markdown */ const fs = require('fs'); const path = require('path'); // JSX component conversion rules const conversions = { // Frame component - extract content and caption frame: (content, caption) => { const cleaned = content.trim(); return caption ? `\n${cleaned}\n\n*${caption}*\n` : `\n${cleaned}\n`; }, // CardGroup and Card components - convert to list cardGroup: (content) => { const cards = []; const cardRegex = /]*href="([^"]+)"[^>]*>([\s\S]*?)<\/Card>/g; let match; while ((match = cardRegex.exec(content)) !== null) { const [, title, href, description] = match; cards.push(`- **[${title}](${href})**: ${description.trim()}`); } return cards.length ? '\n' + cards.join('\n') + '\n' : ''; }, // Tabs component - convert to sections tabs: (content) => { const tabs = []; const tabRegex = /([\s\S]*?)<\/Tab>/g; let match; while ((match = tabRegex.exec(content)) !== null) { const [, title, tabContent] = match; // Clean up the content: remove excessive indentation and whitespace const cleaned = tabContent.trim().replace(/^[ \t]{4,}/gm, ''); tabs.push(`### ${title}\n\n${cleaned}\n`); } return tabs.length ? '\n' + tabs.join('\n') + '\n' : ''; }, // Info/Warning/Tip components - convert to blockquotes note: (content, props) => { const type = props.type || 'note'; const icon = { info: 'ℹ️', warning: '⚠️', tip: '💡' }[type] || 'ℹ️'; return `\n> ${icon} ${content.trim()}\n`; }, // Image components - convert to markdown images img: (props) => { const src = props.src || ''; const alt = props.alt || ''; // Only include light mode images, skip dark mode variants if (props.className && (props.className.includes('dark:') || props.className.includes('hidden'))) { return ''; } // Add newline before and after image for proper spacing return `\n![${alt}](${src})\n`; }, // FAQ/FAQItem components - convert to H2 + H3 Q&A markdown faq: (content) => { const items = []; const itemRegex = /]*?question="([^"]+)"[^>]*>([\s\S]*?)<\/FAQItem>/g; let match; while ((match = itemRegex.exec(content)) !== null) { const [, question, answer] = match; const cleaned = answer.trim().replace(/^[ \t]{2,}/gm, ''); items.push(`### ${question}\n\n${cleaned}\n`); } return items.length ? '\n## Common Questions\n\n' + items.join('\n') + '\n' : ''; }, // Accordion components - convert to H3 + body accordion: (content, title) => { const cleaned = content.trim().replace(/^[ \t]{2,}/gm, ''); return `\n### ${title}\n\n${cleaned}\n`; }, }; // Mask fenced code blocks and inline code so subsequent regex passes (JSX // stripping, orphan-tag sweep) never see their contents. Uppercase-starting // tokens in angle brackets inside shell heredocs (e.g. `<`) look // identical to JSX to our strippers and would otherwise get mangled. function maskCodeSpans(text) { const spans = []; const placeholder = (i) => `\u0000CODESPAN${i}\u0000`; // Fenced code blocks first so inline-code regex can't chew into them. let masked = text.replace(/```[\s\S]*?```/g, (m) => { spans.push(m); return placeholder(spans.length - 1); }); masked = masked.replace(/`[^`\n]*`/g, (m) => { spans.push(m); return placeholder(spans.length - 1); }); return { masked, spans }; } function unmaskCodeSpans(text, spans) { return text.replace(/\u0000CODESPAN(\d+)\u0000/g, (_, i) => spans[Number(i)]); } function convertMdxToMarkdown(mdxContent) { let markdown = mdxContent; // Remove frontmatter but keep title/description as header markdown = markdown.replace(/^---\n([\s\S]*?)\n---\n/, (match, frontmatter) => { const titleMatch = frontmatter.match(/title:\s*(.+)/); const descMatch = frontmatter.match(/description:\s*(.+)/); let header = ''; if (titleMatch) { header += `# ${titleMatch[1]}\n\n`; } if (descMatch) { header += `> ${descMatch[1]}\n\n`; } return header; }); // Convert Frame components (extract content and caption) markdown = markdown.replace(/]*)>([\s\S]*?)<\/Frame>/g, (match, attrs, content) => { const captionMatch = attrs.match(/caption="([^"]*)"/); const caption = captionMatch ? captionMatch[1] : null; return conversions.frame(content, caption); }); // Convert CardGroup components markdown = markdown.replace(/]*>([\s\S]*?)<\/CardGroup>/g, (match, content) => { return conversions.cardGroup(content); }); // Convert Tabs components markdown = markdown.replace(/]*>([\s\S]*?)<\/Tabs>/g, (match, content) => { return conversions.tabs(content); }); // Convert FAQ components markdown = markdown.replace(/]*>([\s\S]*?)<\/FAQ>/g, (match, content) => { return conversions.faq(content); }); // Convert Accordion components (inside or outside AccordionGroup). // Non-greedy body matches the first , so nested Accordions // (rare in our docs) would leave an orphan tag - caught by the // belt-and-suspenders strip below. markdown = markdown.replace(/]*?title="([^"]+)"[^>]*>([\s\S]*?)<\/Accordion>/g, (match, title, content) => { return conversions.accordion(content, title); }); // Unwrap AccordionGroup wrappers (inner Accordions are already converted above) markdown = markdown.replace(/]*>([\s\S]*?)<\/AccordionGroup>/g, (match, content) => content); // Strip the agent-facing inline directive (only useful in HTML, not in .md mirror). // Matches the visually-hidden div added at the top of every page. markdown = markdown.replace(/

]*?>For LLM agents:[\s\S]*?<\/div>/g, ''); // Legacy snippet form, kept for any straggler files markdown = markdown.replace(//g, ''); // Strip snippet import statements (they don't belong in .md output). // Restricted to .js/.jsx so non-JS snippet paths (CSS, etc.) are not stripped. markdown = markdown.replace(/^import\s+\{[^}]*\}\s+from\s+['"]\/snippets\/[^'"]+\.jsx?['"];?\s*$/gm, ''); // Convert Info/Note/Warning components markdown = markdown.replace(/<(Info|Note|Warning|Tip)[^>]*>([\s\S]*?)<\/\1>/g, (match, tag, content) => { return conversions.note(content, { type: tag.toLowerCase() }); }); // Convert img tags markdown = markdown.replace(/]+)\s*\/?>/g, (match, attrs) => { const props = {}; const attrRegex = /(\w+)="([^"]*)"/g; let attrMatch; while ((attrMatch = attrRegex.exec(attrs)) !== null) { props[attrMatch[1]] = attrMatch[2]; } return conversions.img(props); }); // Mask fenced code blocks and inline code so the generic JSX strippers // below never see shell heredocs like `<>` (they look like JSX and // get chewed). Component-specific conversions above run BEFORE masking so // they can still dedent their own code-fenced children correctly. const { masked, spans } = maskCodeSpans(markdown); markdown = masked; // Remove any remaining JSX components. Iterate so nested components // (e.g. ...) fully unwind - a single non-greedy // pass matches the inner pair and leaves the outer closing tag orphaned. const pairRegex = /<[A-Z][^>]*>[\s\S]*?<\/[A-Z][^>]*>/g; let prev; do { prev = markdown; markdown = markdown.replace(pairRegex, ''); } while (markdown !== prev); markdown = markdown.replace(/<[A-Z][^>]*\/>/g, ''); // Belt-and-suspenders: strip any stray orphan JSX tags so Mintlify's .md // parser never trips on "Unexpected closing slash / in tag". markdown = markdown.replace(/<\/?[A-Z][^>]*>/g, ''); // Remove className attributes from remaining tags markdown = markdown.replace(/\s+className="[^"]*"/g, ''); // Clean up excessive newlines (3+ blank lines → 2 blank lines) markdown = markdown.replace(/\n{4,}/g, '\n\n\n'); // Remove lines that are just whitespace markdown = markdown.replace(/^\s+$/gm, ''); // Clean up multiple consecutive blank lines again after whitespace removal markdown = markdown.replace(/\n{3,}/g, '\n\n'); return unmaskCodeSpans(markdown, spans).trim() + '\n'; } function processFile(filePath) { const mdxContent = fs.readFileSync(filePath, 'utf8'); const markdownContent = convertMdxToMarkdown(mdxContent); // Write to .md file (replace .mdx extension) const mdPath = filePath.replace(/\.mdx$/, '.md'); fs.writeFileSync(mdPath, markdownContent, 'utf8'); console.log(`✓ Converted ${path.basename(filePath)} → ${path.basename(mdPath)}`); } function processDirectory(dir) { const entries = fs.readdirSync(dir, { withFileTypes: true }); for (const entry of entries) { const fullPath = path.join(dir, entry.name); if (entry.isDirectory() && !entry.name.startsWith('.')) { processDirectory(fullPath); } else if (entry.isFile() && entry.name.endsWith('.mdx')) { processFile(fullPath); } } } // Main execution const docsDir = __dirname; console.log(`Converting MDX to Markdown in ${docsDir}...\n`); processDirectory(docsDir); console.log('\n✓ All MDX files converted to clean markdown');