Render Markdown

Setup

Install the framework-neutral package:

pnpm add @tanstack/markdown

Render supported Markdown to HTML:

import { renderHtml } from '@tanstack/markdown/html'

const source = `# Release notes

- Small browser bundle
- Serializable AST
- Safe HTML defaults`

const html = renderHtml(source)

console.log(html)

Use the narrow @tanstack/markdown/parser and @tanstack/markdown/html entry points when the default entry's combined exports are unnecessary.

Core Patterns

Parse once and render a reusable document

import { renderHtml } from '@tanstack/markdown/html'
import { parseMarkdown } from '@tanstack/markdown/parser'
import type { MarkdownDocument } from '@tanstack/markdown'

const source = `# Guide

Read the [installation notes](/installation).`

const document = parseMarkdown(source)
const serialized = JSON.stringify(document)
const restored = JSON.parse(serialized) as MarkdownDocument
const html = renderHtml(restored)

console.log(html)

All complete-document renderers accept a source string or an existing MarkdownDocument; a document input skips parsing.

Configure frontmatter and heading IDs while parsing

import { renderHtml } from '@tanstack/markdown/html'
import { parseMarkdown } from '@tanstack/markdown/parser'

const source = `---
title: Installation
published: true
---

# Install

## Install`

const document = parseMarkdown(source, {
  frontmatter: true,
  headingIds(text, lineIndex) {
    const slug = text.toLowerCase().replaceAll(' ', '-')
    return `docs-${lineIndex}-${slug}`
  },
})

console.log(document.frontmatter)
console.log(renderHtml(document))

Frontmatter remains an unparsed string. Heading IDs are generated during parsing and are duplicate-safe under the default slugger.

Add visible heading anchors while rendering

import { renderHtml } from '@tanstack/markdown/html'
import { parseMarkdown } from '@tanstack/markdown/parser'

const document = parseMarkdown('# API')

const html = renderHtml(document, {
  headingAnchors: {
    content: '#',
    className: 'heading-anchor',
    ariaHidden: true,
    tabIndex: -1,
  },
})

console.log(html)

Heading IDs are a parse concern; visible anchor links are a render concern.

Parse inline content only when no block context is needed

import { parseInline, renderInline } from '@tanstack/markdown'

const nodes = parseInline('Use **stable** APIs and `parseMarkdown`.')
const html = nodes.map(node => renderInline(node)).join('')

console.log(html)

parseInline handles inline syntax only. Use parseMarkdown for headings, lists, tables, frontmatter, reference definitions, and footnote definitions.

Behavioral Contracts

• Parsing is synchronous, deterministic, and normalizes line endings.
• Raw block and inline HTML are recognized only with allowHtml: true.
• Link and image URLs have unsafe executable protocols removed.
• Tight lists place simple content directly under <li>; loose lists retain

paragraph wrappers; task checkboxes remain inline with labels.

• Reference definitions resolve case-insensitively before block parsing.
• Footnotes render in first-reference order with collision-safe IDs and

repeated-reference back links.

• Code fence metadata is recorded in the AST; highlighting is external.
• The AST is public but pre-1.0. Regenerate persisted documents when an

upgrade changes node contracts.

Compatibility and Option Timing

TanStack Markdown targets controlled blog and documentation content rather than complete CommonMark, GFM, MDX, or arbitrary HTML parsing. Before adding syntax, require corpus evidence, regression coverage, renderer parity, and an accepted bundle cost. See production-pipelines/SKILL.md for compatibility, security, highlighting, and size gates.

Parsing fixes the document structure. Apply frontmatter, headingIds, allowHtml, parser state, and parser/transform extensions before caching the AST. Renderer-only options such as headingAnchors, highlighter, and codeLineNumbers may be applied when rendering an existing document.

Common Mistakes

HIGH Assuming complete CommonMark or GFM behavior

Wrong:

import { renderHtml } from '@tanstack/markdown/html'

const markdownFromAnywhere = `Heading
===

Visit https://example.com`

const html = renderHtml(markdownFromAnywhere)

console.log(html)

Correct:

import { renderHtml } from '@tanstack/markdown/html'

const controlledMarkdown = `# Heading

Visit [example](https://example.com)`

const html = renderHtml(controlledMarkdown)

console.log(html)

Setext headings and automatic URL linking are outside the supported syntax profile, so unsupported input can remain literal or have different structure.

Source: docs/core-concepts/syntax-profile.md

See also: production-pipelines/SKILL.md - compatibility breadth must be justified against corpus evidence and the bundle budget.

MEDIUM Reparsing unchanged content on every render

Wrong:

import { renderHtml } from '@tanstack/markdown/html'

const source = '# Cached article'

function renderArticle() {
  return renderHtml(source)
}

console.log(renderArticle())
console.log(renderArticle())

Correct:

import { renderHtml } from '@tanstack/markdown/html'
import { parseMarkdown } from '@tanstack/markdown/parser'

const source = '# Cached article'
const document = parseMarkdown(source)

function renderArticle() {
  return renderHtml(document)
}

console.log(renderArticle())
console.log(renderArticle())

Complete-document renderers parse string inputs on every call, while a MarkdownDocument input skips repeated parsing.

Source: docs/core-concepts/document-model.md

See also: production-pipelines/SKILL.md - parse-ahead caching must preserve the parser option and extension set.

HIGH Applying parser options after parsing

Wrong:

import { renderHtml } from '@tanstack/markdown/html'
import { parseMarkdown } from '@tanstack/markdown/parser'

const document = parseMarkdown('# API')
const html = renderHtml(document, { headingIds: false })

console.log(html)

Correct:

import { renderHtml } from '@tanstack/markdown/html'
import { parseMarkdown } from '@tanstack/markdown/parser'

const document = parseMarkdown('# API', { headingIds: false })
const html = renderHtml(document)

console.log(html)

Parse-only options do not retroactively alter a pre-parsed AST.

Source: docs/reference/html.md

See also: custom-extensions/SKILL.md - parser and transform extensions must also be present while creating a cached document.

MEDIUM Using parseInline for document definitions

Wrong:

import { parseInline, renderInline } from '@tanstack/markdown'

const source = `[Guide][guide]

[guide]: /guide`

const html = parseInline(source).map(node => renderInline(node)).join('')

console.log(html)

Correct:

import { renderHtml } from '@tanstack/markdown/html'
import { parseMarkdown } from '@tanstack/markdown/parser'

const source = `[Guide][guide]

[guide]: /guide`

const document = parseMarkdown(source)
const html = renderHtml(document)

console.log(html)

Standalone inline parsing does not extract document-level reference or footnote definitions unless internal parser state is supplied explicitly.

Source: docs/reference/default-entry.md

Related Skills

• production-pipelines/SKILL.md - trust boundaries, syntax highlighting,

practical compatibility, caching, tests, performance, and bundle budgets.

• react-rendering/SKILL.md - render the same source or AST as React nodes.
• octane-rendering/SKILL.md - render the same source or AST as Octane nodes.
• custom-extensions/SKILL.md - add deterministic parser and renderer hooks.
• docs-features/SKILL.md - use the first-party documentation extensions.

References

• [AST nodes, parser state, and render options](references/ast-and-options.md)