Markdown Abstract Syntax Tree.
mdast is a specification for representing Markdown in a syntax tree. It implements the unist spec. It can represent several flavours of Markdown, such as CommonMark and GitHub Flavored Markdown.
This document may not be released.
See releases for released documents.
The latest released version is 4.0.0
.
- Introduction
- Nodes
- Mixin
- Enumeration
- Content model
- Extensions
- Glossary
- List of utilities
- References
- Security
- Related
- Contribute
- Acknowledgments
- License
This document defines a format for representing Markdown as an abstract syntax tree. Development of mdast started in July 2014, in remark, before unist existed. This specification is written in a Web IDL-like grammar.
mdast extends unist, a format for syntax trees, to benefit from its ecosystem of utilities.
mdast relates to JavaScript in that it has a rich ecosystem of utilities for working with compliant syntax trees in JavaScript. However, mdast is not limited to JavaScript and can be used in other programming languages.
mdast relates to the unified and remark projects in that mdast syntax trees are used throughout their ecosystems.
interface Parent <: UnistParent {
children: [MdastContent]
}
Parent (UnistParent) represents a node in mdast containing other nodes (said to be children).
Its content is limited to only other mdast content.
interface Literal <: UnistLiteral {
value: string
}
Literal (UnistLiteral) represents a node in mdast containing a value.
Its value
field is a string
.
interface Root <: Parent {
type: "root"
}
Root (Parent) represents a document.
Root can be used as the root of a tree, never as a child. Its content model is not limited to flow content, but can contain any mdast content with the restriction that all content must be of the same category.
interface Paragraph <: Parent {
type: "paragraph"
children: [PhrasingContent]
}
Paragraph (Parent) represents a unit of discourse dealing with a particular point or idea.
Paragraph can be used where content is expected. Its content model is phrasing content.
For example, the following Markdown:
Alpha bravo charlie.
Yields:
{
type: 'paragraph',
children: [{type: 'text', value: 'Alpha bravo charlie.'}]
}
interface Heading <: Parent {
type: "heading"
depth: 1 <= number <= 6
children: [PhrasingContent]
}
Heading (Parent) represents a heading of a section.
Heading can be used where flow content is expected. Its content model is phrasing content.
A depth
field must be present.
A value of 1
is said to be the highest rank and 6
the lowest.
For example, the following Markdown:
# Alpha
Yields:
{
type: 'heading',
depth: 1,
children: [{type: 'text', value: 'Alpha'}]
}
interface ThematicBreak <: Node {
type: "thematicBreak"
}
ThematicBreak (Node) represents a thematic break, such as a scene change in a story, a transition to another topic, or a new document.
ThematicBreak can be used where flow content is expected. It has no content model.
For example, the following Markdown:
***
Yields:
{type: 'thematicBreak'}
interface Blockquote <: Parent {
type: "blockquote"
children: [FlowContent]
}
Blockquote (Parent) represents a section quoted from somewhere else.
Blockquote can be used where flow content is expected. Its content model is also flow content.
For example, the following Markdown:
> Alpha bravo charlie.
Yields:
{
type: 'blockquote',
children: [{
type: 'paragraph',
children: [{type: 'text', value: 'Alpha bravo charlie.'}]
}]
}
interface List <: Parent {
type: "list"
ordered: boolean?
start: number?
spread: boolean?
children: [ListContent]
}
List (Parent) represents a list of items.
List can be used where flow content is expected. Its content model is list content.
An ordered
field can be present.
It represents that the items have been intentionally ordered (when true
), or
that the order of items is not important (when false
or not present).
A start
field can be present.
It represents, when the ordered
field is true
, the starting number of the
list.
A spread
field can be present.
It represents that one or more of its children are separated with a blank line
from its siblings (when true
), or not (when false
or not
present).
For example, the following Markdown:
1. foo
Yields:
{
type: 'list',
ordered: true,
start: 1,
spread: false,
children: [{
type: 'listItem',
spread: false,
children: [{
type: 'paragraph',
children: [{type: 'text', value: 'foo'}]
}]
}]
}
interface ListItem <: Parent {
type: "listItem"
spread: boolean?
children: [FlowContent]
}
ListItem (Parent) represents an item in a List.
ListItem can be used where list content is expected. Its content model is flow content.
A spread
field can be present.
It represents that the item contains two or more children
separated by a blank line (when true
), or not (when false
or not present).
For example, the following Markdown:
* bar
Yields:
{
type: 'listItem',
spread: false,
children: [{
type: 'paragraph',
children: [{type: 'text', value: 'bar'}]
}]
}
interface HTML <: Literal {
type: "html"
}
HTML (Literal) represents a fragment of raw HTML.
HTML can be used where flow or
phrasing content is expected.
Its content is represented by its value
field.
HTML nodes do not have the restriction of being valid or complete HTML ([HTML]) constructs.
For example, the following Markdown:
<div>
Yields:
{type: 'html', value: '<div>'}
interface Code <: Literal {
type: "code"
lang: string?
meta: string?
}
Code (Literal) represents a block of preformatted text, such as ASCII art or computer code.
Code can be used where flow content is expected.
Its content is represented by its value
field.
This node relates to the phrasing content concept InlineCode.
A lang
field can be present.
It represents the language of computer code being marked up.
If the lang
field is present, a meta
field can be present.
It represents custom information relating to the node.
For example, the following Markdown:
foo()
Yields:
{
type: 'code',
lang: null,
meta: null,
value: 'foo()'
}
And the following Markdown:
```js highlight-line="2"
foo()
bar()
baz()
```
Yields:
{
type: 'code',
lang: 'javascript',
meta: 'highlight-line="2"',
value: 'foo()\nbar()\nbaz()'
}
interface Definition <: Node {
type: "definition"
}
Definition includes Association
Definition includes Resource
Definition (Node) represents a resource.
Definition can be used where content is expected. It has no content model.
Definition includes the mixins Association and Resource.
Definition should be associated with LinkReferences and ImageReferences.
For example, the following Markdown:
[Alpha]: https://example.com
Yields:
{
type: 'definition',
identifier: 'alpha',
label: 'Alpha',
url: 'https://example.com',
title: null
}
interface Text <: Literal {
type: "text"
}
Text (Literal) represents everything that is just text.
Text can be used where phrasing content is
expected.
Its content is represented by its value
field.
For example, the following Markdown:
Alpha bravo charlie.
Yields:
{type: 'text', value: 'Alpha bravo charlie.'}
interface Emphasis <: Parent {
type: "emphasis"
children: [TransparentContent]
}
Emphasis (Parent) represents stress emphasis of its contents.
Emphasis can be used where phrasing content is expected. Its content model is transparent content.
For example, the following Markdown:
*alpha* _bravo_
Yields:
{
type: 'paragraph',
children: [
{
type: 'emphasis',
children: [{type: 'text', value: 'alpha'}]
},
{type: 'text', value: ' '},
{
type: 'emphasis',
children: [{type: 'text', value: 'bravo'}]
}
]
}
interface Strong <: Parent {
type: "strong"
children: [TransparentContent]
}
Strong (Parent) represents strong importance, seriousness, or urgency for its contents.
Strong can be used where phrasing content is expected. Its content model is transparent content.
For example, the following Markdown:
**alpha** __bravo__
Yields:
{
type: 'paragraph',
children: [
{
type: 'strong',
children: [{type: 'text', value: 'alpha'}]
},
{type: 'text', value: ' '},
{
type: 'strong',
children: [{type: 'text', value: 'bravo'}]
}
]
}
interface InlineCode <: Literal {
type: "inlineCode"
}
InlineCode (Literal) represents a fragment of computer code, such as a file name, computer program, or anything a computer could parse.
InlineCode can be used where phrasing content
is expected.
Its content is represented by its value
field.
This node relates to the flow content concept Code.
For example, the following Markdown:
`foo()`
Yields:
{type: 'inlineCode', value: 'foo()'}
interface Break <: Node {
type: "break"
}
Break (Node) represents a line break, such as in poems or addresses.
Break can be used where phrasing content is expected. It has no content model.
For example, the following Markdown:
foo··
bar
Yields:
{
type: 'paragraph',
children: [
{type: 'text', value: 'foo'},
{type: 'break'},
{type: 'text', value: 'bar'}
]
}
interface Link <: Parent {
type: "link"
children: [StaticPhrasingContent]
}
Link includes Resource
Link (Parent) represents a hyperlink.
Link can be used where phrasing content is expected. Its content model is static phrasing content.
Link includes the mixin Resource.
For example, the following Markdown:
[alpha](https://example.com "bravo")
Yields:
{
type: 'link',
url: 'https://example.com',
title: 'bravo',
children: [{type: 'text', value: 'alpha'}]
}
interface Image <: Node {
type: "image"
}
Image includes Resource
Image includes Alternative
Image (Node) represents an image.
Image can be used where phrasing content is
expected.
It has no content model, but is described by its alt
field.
Image includes the mixins Resource and Alternative.
For example, the following Markdown:
![alpha](https://example.com/favicon.ico "bravo")
Yields:
{
type: 'image',
url: 'https://example.com/favicon.ico',
title: 'bravo',
alt: 'alpha'
}
interface LinkReference <: Parent {
type: "linkReference"
children: [StaticPhrasingContent]
}
LinkReference includes Reference
LinkReference (Parent) represents a hyperlink through association, or its original source if there is no association.
LinkReference can be used where phrasing content is expected. Its content model is static phrasing content.
LinkReference includes the mixin Reference.
LinkReferences should be associated with a Definition.
For example, the following Markdown:
[alpha][Bravo]
Yields:
{
type: 'linkReference',
identifier: 'bravo',
label: 'Bravo',
referenceType: 'full',
children: [{type: 'text', value: 'alpha'}]
}
interface ImageReference <: Node {
type: "imageReference"
}
ImageReference includes Reference
ImageReference includes Alternative
ImageReference (Node) represents an image through association, or its original source if there is no association.
ImageReference can be used where phrasing
content is expected.
It has no content model, but is described by its alt
field.
ImageReference includes the mixins Reference and Alternative.
ImageReference should be associated with a Definition.
For example, the following Markdown:
![alpha][bravo]
Yields:
{
type: 'imageReference',
identifier: 'bravo',
label: 'bravo',
referenceType: 'full',
alt: 'alpha'
}
interface mixin Resource {
url: string
title: string?
}
Resource represents a reference to resource.
A url
field must be present.
It represents a URL to the referenced resource.
A title
field can be present.
It represents advisory information for the resource, such as would be
appropriate for a tooltip.
interface mixin Association {
identifier: string
label: string?
}
Association represents an internal relation from one node to another.
An identifier
field must be present.
It can match another node.
identifier
is a source value: character escapes and character references are
not parsed.
Its value must be normalized.
A label
field can be present.
label
is a string value: it works just like title
on a link or a lang
on
code: character escapes and character references are parsed.
To normalize a value, collapse markdown whitespace ([\t\n\r ]+
) to a space,
trim the optional initial and/or final space, and perform case-folding.
Whether the value of identifier
(or normalized label
if there is no
identifier
) is expected to be a unique identifier or not depends on the type
of node including the Association.
An example of this is that they should be unique on
Definition, whereas multiple
LinkReferences can be non-unique to be associated with
one definition.
interface mixin Reference {
referenceType: string
}
Reference includes Association
Reference represents a marker that is associated to another node.
A referenceType
field must be present.
Its value must be a referenceType.
It represents the explicitness of the reference.
interface mixin Alternative {
alt: string?
}
Alternative represents a node with a fallback
An alt
field should be present.
It represents equivalent content for environments that cannot represent the
node as intended.
enum referenceType {
"shortcut" | "collapsed" | "full"
}
referenceType represents the explicitness of a reference.
- shortcut: the reference is implicit, its identifier inferred from its content
- collapsed: the reference is explicit, its identifier inferred from its content
- full: the reference is explicit, its identifier explicitly set
type MdastContent = FlowContent | ListContent | PhrasingContent
Each node in mdast falls into one or more categories of Content that group nodes with similar characteristics together.
type FlowContent =
Blockquote | Code | Heading | HTML | List | ThematicBreak | Content
Flow content represent the sections of document.
type Content = Definition | Paragraph
Content represents runs of text that form definitions and paragraphs.
type ListContent = ListItem
List content represent the items in a list.
type PhrasingContent = Link | LinkReference | StaticPhrasingContent
Phrasing content represent the text in a document, and its markup.
type StaticPhrasingContent =
Break | Emphasis | HTML | Image | ImageReference | InlineCode | Strong | Text
StaticPhrasing content represent the text in a document, and its markup, that is not intended for user interaction.
The transparent content model is derived from the content model of its parent. Effectively, this is used to prohibit nested links (and link references).
Markdown syntax is often extended. It is not a goal of this specification to list all possible extensions. However, a short list of frequently used extensions are shown below.
The following interfaces are found in GitHub Flavored Markdown.
interface Table <: Parent {
type: "table"
align: [alignType]?
children: [TableContent]
}
Table (Parent) represents two-dimensional data.
Table can be used where flow content is expected. Its content model is table content.
The head of the node represents the labels of the columns.
An align
field can be present.
If present, it must be a list of alignTypes.
It represents how cells in columns are aligned.
For example, the following Markdown:
| foo | bar |
| :-- | :-: |
| baz | qux |
Yields:
{
type: 'table',
align: ['left', 'center'],
children: [
{
type: 'tableRow',
children: [
{
type: 'tableCell',
children: [{type: 'text', value: 'foo'}]
},
{
type: 'tableCell',
children: [{type: 'text', value: 'bar'}]
}
]
},
{
type: 'tableRow',
children: [
{
type: 'tableCell',
children: [{type: 'text', value: 'baz'}]
},
{
type: 'tableCell',
children: [{type: 'text', value: 'qux'}]
}
]
}
]
}
interface TableRow <: Parent {
type: "tableRow"
children: [RowContent]
}
TableRow (Parent) represents a row of cells in a table.
TableRow can be used where table content is expected. Its content model is row content.
If the node is a head, it represents the labels of the columns for its parent Table.
For an example, see Table.
interface TableCell <: Parent {
type: "tableCell"
children: [PhrasingContent]
}
TableCell (Parent) represents a header cell in a Table, if its parent is a head, or a data cell otherwise.
TableCell can be used where row content is expected. Its content model is phrasing content excluding Break nodes.
For an example, see Table.
interface ListItemGfm <: ListItem {
checked: boolean?
}
In GFM, a checked
field can be present.
It represents whether the item is done (when true
), not done (when false
),
or indeterminate or not applicable (when null
or not present).
interface Delete <: Parent {
type: "delete"
children: [TransparentContent]
}
Delete (Parent) represents contents that are no longer accurate or no longer relevant.
Delete can be used where phrasing content is expected. Its content model is transparent content.
For example, the following Markdown:
~~alpha~~
Yields:
{
type: 'delete',
children: [{type: 'text', value: 'alpha'}]
}
enum alignType {
"left" | "right" | "center" | null
}
alignType represents how phrasing content is aligned ([CSSTEXT]).
'left'
: See theleft
value of thetext-align
CSS property'right'
: See theright
value of thetext-align
CSS property'center'
: See thecenter
value of thetext-align
CSS propertynull
: phrasing content is aligned as defined by the host environment
type FlowContentGfm = Table | FlowContent
type TableContent = TableRow
Table content represent the rows in a table.
type RowContent = TableCell
Row content represent the cells in a row.
type ListContentGfm = ListItemGfm
type StaticPhrasingContentGfm = Delete | StaticPhrasingContent
The following interfaces are found with YAML.
interface YAML <: Literal {
type: "yaml"
}
YAML (Literal) represents a collection of metadata for the document in the YAML ([YAML]) data serialisation language.
YAML can be used where frontmatter content is
expected.
Its content is represented by its value
field.
For example, the following Markdown:
---
foo: bar
---
Yields:
{type: 'yaml', value: 'foo: bar'}
type FrontmatterContent = YAML
type FlowContentFrontmatter = FrontmatterContent | FlowContent
Frontmatter content represent out-of-band information about the document.
If frontmatter is present, it must be limited to one node in the tree, and can only exist as a head.
type FlowContentFrontmatter = FrontmatterContent | FlowContent
The following interfaces are found with footnotes.
interface FootnoteDefinition <: Parent {
type: "footnoteDefinition"
children: [FlowContent]
}
FootnoteDefinition includes Association
FootnoteDefinition (Parent) represents content relating to the document that is outside its flow.
FootnoteDefinition can be used where flow content is expected. Its content model is also flow content.
FootnoteDefinition includes the mixin Association.
FootnoteDefinition should be associated with FootnoteReferences.
For example, the following Markdown:
[^alpha]: bravo and charlie.
Yields:
{
type: 'footnoteDefinition',
identifier: 'alpha',
label: 'alpha',
children: [{
type: 'paragraph',
children: [{type: 'text', value: 'bravo and charlie.'}]
}]
}
interface Footnote <: Parent {
type: "footnote"
children: [PhrasingContent]
}
Footnote (Parent) represents content relating to the document that is outside its flow.
Footnote can be used where phrasing content is expected. Its content model is also phrasing content.
For example, the following Markdown:
^[alpha bravo]
Yields:
{
type: 'footnote',
children: [{type: 'text', value: 'alpha bravo'}]
}
interface FootnoteReference <: Node {
type: "footnoteReference"
}
FootnoteReference includes Association
FootnoteReference (Node) represents a marker through association.
FootnoteReference can be used where phrasing content is expected. It has no content model.
FootnoteReference includes the mixin Association.
FootnoteReference should be associated with a FootnoteDefinition.
For example, the following Markdown:
[^alpha]
Yields:
{
type: 'footnoteReference',
identifier: 'alpha',
label: 'alpha'
}
type FlowContentFootnotes = FootnoteDefinition | FlowContent
type StaticPhrasingContentFootnotes =
Footnote | FootnoteReference | StaticPhrasingContent
See the unist glossary.
See the unist list of utilities for more utilities.
mdast-add-list-metadata
— Enhances the metadata of list and listItem nodesmdast-util-assert
— Assert nodesmdast-builder
— Build mdast structures with composable functionsmdast-comment-marker
— Parse a comment markermdast-util-compact
— Make a tree compactmdast-util-definitions
— Find definition nodesmdast-util-from-quill-delta
— Transform Quill delta to mdastmdast-flatten-image-paragraphs
— Flatten paragraph and image into one image nodemdast-flatten-listitem-paragraphs
— Flatten listItem and (nested) paragraph into one listItem nodemdast-flatten-nested-lists
— Transforms a tree to avoid lists inside listsmdast-util-heading-range
— Markdown heading as rangesmdast-util-heading-style
— Get the style of a heading nodemdast-util-inject
— Inject a tree into another at a given headingmdast-move-images-to-root
— Moves image nodes up the tree until they are strict children of the rootmdast-normalize-headings
— Ensure at most one top-level heading is in the documentmdast-util-phrasing
— Check if a node is phrasing contentmdast-squeeze-paragraphs
— Remove empty paragraphsmdast-util-toc
— Generate a Table of Contents from a treemdast-util-to-hast
— Transform to hastmdast-util-to-nlcst
— Transform to nlcstmdast-util-to-string
— Get the plain text content of a nodemdast-zone
— HTML comments as ranges or markers
- unist: Universal Syntax Tree. T. Wormer; et al.
- Markdown: Markdown. J. Gruber.
- CommonMark: CommonMark. J. MacFarlane; et al.
- GFM: GitHub Flavored Markdown. GitHub.
- HTML: HTML Standard, A. van Kesteren; et al. WHATWG.
- CSSTEXT: CSS Text, CSS Text, E. Etemad, K. Ishii. W3C.
- JavaScript: ECMAScript Language Specification. Ecma International.
- YAML: YAML Ain’t Markup Language, O. Ben-Kiki, C. Evans, I. döt Net.
- Web IDL: Web IDL, C. McCormack. W3C.
As mdast can contain HTML and be used to represent HTML, and improper use of
HTML can open you up to a cross-site scripting (XSS) attack, improper use
of mdast is also unsafe.
When transforming to HTML (typically through hast), always be
careful with user input and use hast-util-santize
to make the hast
tree safe.
- hast — Hypertext Abstract Syntax Tree format
- nlcst — Natural Language Concrete Syntax Tree format
- xast — Extensible Abstract Syntax Tree
See contributing.md
in syntax-tree/.github
for
ways to get started.
See support.md
for ways to get help.
Ideas for new utilities and tools can be posted in syntax-tree/ideas
.
A curated list of awesome syntax-tree, unist, mdast, hast, xast, and nlcst resources can be found in awesome syntax-tree.
This project has a code of conduct. By interacting with this repository, organization, or community you agree to abide by its terms.
The initial release of this project was authored by @wooorm.
Special thanks to @eush77 for their work, ideas, and incredibly valuable feedback!
Thanks to @anandthakker, @arobase-che, @BarryThePenguin, @chinesedfan, @ChristianMurphy, @craftzdog, @d4rekanguok, @detj, @dominictarr, @gkatsev, @Hamms, @Hypercubed, @ikatyang, @izumin5210, @jasonLaster, @Justineo, @justjake, @KyleAMathews, @laysent, @macklinu, @mike-north, @Murderlon, @nevik, @Rokt33r, @rhysd, @rubys, @Sarah-Seo, @sethvincent, @silvenon, @simov, @staltz, @stefanprobst, @tmcw, and @vhf for contributing to mdast and related projects!