mdlinkcheck

Find broken relative links and #anchors in Markdown files: offline, fast, with no dependencies, and using the same heading-anchor rules as GitHub.

$ mdlinkcheck src/engine
src/engine/prague.md:72:38: #BlobsBundleV1: no heading or anchor '#BlobsBundleV1' in this file [missing-anchor]
src/engine/refactor-ssz.md:23:23: #max-constants: no heading or anchor '#max-constants' in this file; did you mean '#max_-constants'? [missing-anchor]
src/engine/refactor-ssz.md:524:26: #get-forkpayloadspayloadid: no heading or anchor '#get-forkpayloadspayloadid' in this file; did you mean '#get-payloadspayloadid'? [missing-anchor]
src/engine/refactor.md:447:39: ./refactor-ssz.md#max-constants: no heading or anchor '#max-constants' in 'refactor-ssz.md'; did you mean '#max_-constants'? [missing-anchor]
mdlinkcheck: 13 files, 460 links, 4 broken

Why

External-URL checkers (lychee, markdown-link-check) are great at telling you a website is down. They are much less good at the links that break inside a repository:

• a section gets renamed and every other-doc.md#old-title link silently stops scrolling anywhere;
• a heading contains code, punctuation or underscores, so its real anchor isn't what you'd guess (## `MAX_*` constants is #max_-constants, ## Definition of `channel_id` is #definition-of-channel_id);
• a file moves and ../dump.md becomes ./dump.md in one place but not another;
• a link works on your Mac but not on GitHub, because the file name differs only in case.

GitHub doesn't warn about any of these. mdlinkcheck does, without touching the network, so it is fast enough for a pre-commit hook and deterministic enough for CI.

Install

pipx install mdlinkcheck
# or
pip install mdlinkcheck

Python 3.9 or newer; no third-party dependencies.

Usage

mdlinkcheck                 # every Markdown file in the repository
mdlinkcheck docs README.md  # specific files and directories

It exits with 1 if any link is broken, 0 otherwise, and 2 on usage errors.

Docs sites

Docs generators serve pages by route, so their links often leave out .md or start with /. mdlinkcheck looks for docusaurus.config.*, vocs.config.*, .vitepress/ and mkdocs.yml and resolves links inside those sites the way the site will: against the pages folder and static assets folder, with or without the extension, and from the parent folder for index pages. The detected folders are printed in the summary; pass --site-root to set them yourself.

To skip a single link, put <!-- mdlinkcheck-ignore --> on the same line or on the line above it.

GitHub Actions

- uses: actions/checkout@v4
- uses: mikeua/mdlinkcheck@v0.2.0
  with:
    args: --exclude CHANGELOG.md .

Broken links show up as annotations on the pull request diff.

pre-commit

repos:
  - repo: https://github.com/mikeua/mdlinkcheck
    rev: v0.2.0
    hooks:
      - id: mdlinkcheck

What it checks

Links are found in inline links and images, reference definitions ([x]: target) and HTML href/src attributes. Code blocks, code spans and HTML comments are skipped, as are external URLs (https:, mailto: and other schemes), GitHub line anchors (#L10-L20) and anything that looks like a template placeholder.

Anchors follow GitHub's rules: the heading is rendered to plain text, lowercased, stripped of everything except letters, digits, -, _ and spaces, and each space becomes -. Repeated headings get -1, -2, ... suffixes, and links may use the user-content- prefix and percent-encoding. Headings inside list items and block quotes count too.

Bugs found with it

• graphprotocol/graph-node#6740: renamed section, missing file, wrong relative path
• ethereum/execution-apis#943: anchors in the Engine API spec
• lightning/bolts#1364: links between Lightning BOLT specs

Development

python -m venv .venv && . .venv/bin/activate
pip install -e . pytest ruff
pytest
ruff check . && ruff format --check .

License

MIT