tree-sitter-markdown-text

Markdown grammar for tree-sitter, shaped so that its AST lines up with the textlint TxtNode model.

Parses .md (and .markdown, .mdown, .mkd, .mkdn) files into a concrete syntax tree covering the full CommonMark block structure plus common extensions (GFM pipe tables, task lists, GFM alerts, YAML/TOML front matter, Pandoc math and directive blocks, footnotes, MDX JSX). Inline content is surfaced as structured children of the inline wrapper: classified tokens (word_token, numeric_token, identifier_like_token, path_like_token) and punctuation-class nodes (terminator, separator, bracket, operator_like), plus inline structural nodes (emphasis, strong, strikethrough, link, image, autolink, inline_code, html_inline, math_inline, mdx_jsx_inline, footnote_reference).

Features

Block nodes

• Document structure — document, nested section wrappers around ATX headings, paragraph, blank_line (as a first-class node).
• Headings — ATX (#..######) and setext (===/---) with the heading level exposed as a level field on both atx_heading and setext_heading.
• Code blocks — indented code blocks and fenced code blocks (backtick and tilde), with info_string/language children for the GFM language tag.
• Math blocks — Pandoc/GitLab/KaTeX display math ($$…$$) as a dedicated math_block with math_block_delimiter/math_block_content children.
• Lists — unordered (+/-/*) and ordered (1./1)) list markers. GFM task list items are promoted to task_list_item (distinct from list_item), with task_list_marker_checked/task_list_marker_unchecked markers.
• Block quotes and callouts — nested quotes and lazy continuations. A block quote whose first paragraph begins with [!NOTE] / [!TIP] / [!IMPORTANT] / [!WARNING] / [!CAUTION] (or any uppercase-only label) is surfaced as callout with a callout_type field.
• Thematic breaks — ---, ***, ___.
• HTML blocks — all 7 CommonMark HTML block types; block-level HTML comments are aliased to html_comment_block for easy metric extraction.
• MDX JSX blocks — shallow mdx_jsx_block for lines that start with an MDX-style JSX element (<Component ...>, <Component/>, </Component>). Component-style mixed-case names disambiguate from all-caps HTML blocks such as <DIV>.
• Pipe tables — pipe_table with pipe_table_header, pipe_table_delimiter_row, pipe_table_row, pipe_table_cell, pipe_table_align_left/pipe_table_align_right.
• Link reference definitions — link_reference_definition with link_label/link_destination/link_title children.
• Footnote definitions — footnote_definition ([^id]: …) with a footnote_label child.
• Directive blocks — generic container directives (:::name … :::, per remark-directive / MyST / Pandoc fenced divs) as directive_block with directive_block_delimiter/directive_name/directive_block_content children.
• Image blocks — a paragraph consisting of a single block-level image (![alt](dest) on its own line) is surfaced as image_block with link_label/link_destination children.
• Front matter — YAML (--- fenced) as minus_metadata, TOML (+++ fenced) as plus_metadata.

Inline nodes (children of the inline wrapper)

• Classified text tokens — text_span wraps runs of classified tokens: word_token (Unicode alphabetic), numeric_token (integers, decimals, versions), identifier_like_token (camelCase / PascalCase / snake_case), path_like_token (paths with / separators or dotted identifiers).

• Punctuation classes — every punctuation lexeme is classified: terminator (., ?, !, 。, …), separator (,, ;, :), bracket ((, ), [, ], {, }, <, >), operator_like (::, ->, =>, =, +, -, *, /, |, &, and other punctuation).

• Emphasis / strong / strikethrough — emphasis (*…* or _…_), strong (**…** or __…__), strikethrough (~~…~~), each with a _delimiter/_content/_delimiter sub-tree.

• Code spans — inline_code with matched backtick-run delimiters (1 or 2 backticks).

• Links and images — link (inline, full-reference, collapsed-reference, shortcut-reference forms) and image (![alt](dest) or ![alt][ref]). Both expose link_label/link_destination/link_title children.

• Autolinks — autolink with uri or email children for <https://…> and <user@example.com>.

• Raw HTML inline — html_inline with html_open_tag/html_close_tag/html_comment/html_cdata/html_declaration/html_processing_instruction children.

• MDX JSX inline — shallow mdx_jsx_inline with mdx_jsx_open_tag/mdx_jsx_close_tag/mdx_jsx_expression children.

• Inline math — math_inline ($…$) with math_inline_delimiter/math_inline_content children. Disambiguated from math_block ($$…$$).

• Footnote references — footnote_reference ([^id] inside prose) with a footnote_reference_label child.

• Injections query — ships a queries/injections.scm that injects into fenced-code-block info strings, HTML blocks, and front matter.

Example

# Heading

A paragraph with inline content.

- one
- two

```go
func main() {}

Parsed tree (abbreviated):

(document (section (atx_heading level: (atx_h1_marker) heading_content: (inline)) (blank_line) (paragraph (inline)) (blank_line) (list (list_item (list_marker_minus) (paragraph (inline))) (list_item (list_marker_minus) (paragraph (inline)))) (blank_line) (fenced_code_block (fenced_code_block_delimiter) (info_string (language)) (code_fence_content) (fenced_code_block_delimiter))))

## Relationship to textlint

The grammar is structurally close to the textlint AST. Every block-level `TxtNode` type has a direct counterpart here; inline `TxtNode` types (`Str`, `Emphasis`, `Strong`, `Link`, `Image`, `Code`, `Html`, `Delete`, `FootnoteReference`) also have direct counterparts as children of the `inline` wrapper. Names stay snake_case per the tree-sitter convention; consumers map names themselves. See [docs/textlint-mapping.md](docs/textlint-mapping.md) for the full table.

## Installation

### npm

```sh
npm install tree-sitter-markdown-text

Cargo

cargo add tree-sitter-markdown-text

PyPI

pip install tree-sitter-markdown-text

Go

import tree_sitter_markdown_text "github.com/ophidiarium/tree-sitter-markdown-text/bindings/go"

The root package also exports the bundled queries via go:embed:

import markdown "github.com/ophidiarium/tree-sitter-markdown-text"

lang := markdown.GetLanguage()
query, _ := markdown.GetHighlightsQuery()

Usage

Node.js

import Parser from "tree-sitter";
import Markdown from "tree-sitter-markdown-text";

const parser = new Parser();
parser.setLanguage(Markdown);

const tree = parser.parse("# hello\n");
console.log(tree.rootNode.toString());

Rust

let mut parser = tree_sitter::Parser::new();
let language = tree_sitter_markdown_text::LANGUAGE;
parser.set_language(&language.into()).unwrap();

let tree = parser.parse("# hello\n", None).unwrap();
println!("{}", tree.root_node().to_sexp());

Python

from tree_sitter import Language, Parser
import tree_sitter_markdown_text

parser = Parser(Language(tree_sitter_markdown_text.language()))
tree = parser.parse(b"# hello\n")
print(tree.root_node.sexp())

Credits and references

• tree-sitter-grammars/tree-sitter-markdown — upstream grammar, specifically the split_parser branch's block grammar, which this grammar is derived from.
• textlint TxtNode — the AST shape this grammar targets for compatibility.
• CommonMark Spec — the block structure this grammar implements.
• Github Flavored Markdown — for the pipe-table and task-list extensions.

License

MIT