dw2md

Grab an entire DeepWiki and compile it into a single markdown file — ready to drop into an LLM context window.

dw2md tinygrad/tinygrad -o tinygrad-wiki.md

DeepWiki generates excellent structured documentation for open-source repositories, but it's spread across dozens of client-rendered pages with no export button. dw2md talks directly to DeepWiki's MCP server to pull the full wiki structure and contents, then assembles everything into one clean document.

Install

From crates.io (recommended)

cargo install dw2md

Homebrew (macOS/Linux)

Coming soon! Homebrew tap is in progress.

From source

git clone https://github.com/tnguyen21/dw2md
cd dw2md
cargo install --path .

Pre-built binaries

Download from GitHub Releases for Linux, macOS (x86_64/ARM64), and Windows.

Produces a single static binary (~6MB, no OpenSSL dependency).

Usage

dw2md [OPTIONS] <REPO>

<REPO> accepts any of:

Options

Examples

Dump an entire wiki to a file:

dw2md tokio-rs/tokio -o tokio-wiki.md

Pipe straight to clipboard (macOS):

dw2md tinygrad/tinygrad | pbcopy

See what sections are available before downloading:

dw2md tinygrad/tinygrad --list

├── 1 Overview [1-overview]
│   └── 1.1 Repository Structure and Packages [1-1-repository-structure-and-packages]
├── 2 Feature Flags System [2-feature-flags-system]
├── 3 Build System and Package Distribution [3-build-system-and-package-distribution]
│   ...
└── 7 Developer Tools and Debugging [7-developer-tools-and-debugging]

Interactively pick which sections to include:

dw2md tinygrad/tinygrad -i -o react-wiki.md

Shows a multi-select prompt where you can toggle sections on/off with space, then press enter to fetch only what you selected. All sections are selected by default.

Only grab specific sections (if you know the slugs):

dw2md tinygrad/tinygrad -p "4-react-reconciler,4-1-fiber-architecture-and-data-structures"

Exclude sections you don't need:

dw2md tinygrad/tinygrad -x "7-developer-tools-and-debugging"

JSON output for programmatic use:

dw2md tinygrad/tinygrad -f json -o react.json

From a DeepWiki URL you already have open:

dw2md https://deepwiki.com/anthropics/claude-code

Minimal output — no metadata, no TOC, just content:

dw2md tinygrad/tinygrad --no-toc --no-metadata

Output

Markdown (default)

The output format is designed for LLM and agent workflows — a tree-structured table of contents for fast orientation, and grep-friendly section delimiters so agents can selectively extract the sections they need.

<!-- dw2md v0.2.0 | tinygrad/tinygrad | 2026-02-12T15:30:00Z | 29 pages -->

# tinygrad/tinygrad — DeepWiki

> Compiled from https://deepwiki.com/tinygrad/tinygrad
> Generated: 2026-02-12T15:30:00Z | Pages: 29

## Structure

├── 1 Overview
│   └── 1.1 Repository Structure and Packages
├── 2 Feature Flags System
├── 3 Build System and Package Distribution
│   ├── 3.1 Setup.py and Package Config
│   └── 3.2 CI/CD Pipeline
└── 4 Core Architecture

## Contents

<<< SECTION: 1 Overview [1-overview] >>>

[page content with original heading levels preserved]

<<< SECTION: 1.1 Repository Structure and Packages [1-1-repository-structure-and-packages] >>>

[page content]

<<< SECTION: 2 Feature Flags System [2-feature-flags-system] >>>

[page content]

Why this format?

The <<< SECTION: Title [slug] >>> delimiter is designed to be trivially grep-able:

# List all sections in a compiled wiki
grep "^<<< SECTION:" wiki.md

# Extract a specific section (content between two delimiters)
sed -n '/^<<< SECTION: 1 Overview/,/^<<< SECTION:/p' wiki.md

# Regex captures both title and slug in one match
# ^<<< SECTION: (.+?) \[(.+?)\] >>>$

This matters because LLMs and agents working with large documents need to scan structure cheaply, then pull in only the sections relevant to their current task — rather than stuffing the entire document into context.

Other format choices:

• Tree TOC — ├──/└── characters show hierarchy at a glance (same as tree command), scannable faster than indented bullet lists
• Original heading levels preserved — no heading-level bumping; the delimiter handles section boundaries, so page content keeps its source structure
• Token efficient — no repeated horizontal rules (---), no anchor link markup, no extra # characters from heading bumping
• HTML comment on line 1 — machine-parseable metadata invisible to most renderers
• Mermaid blocks preserved as fenced code blocks
• Source annotations preserved (Sources: file.py:1-50) for code location context

JSON

With --format json:

{
  "repo": "tinygrad/tinygrad",
  "url": "https://deepwiki.com/tinygrad/tinygrad",
  "generated_at": "2026-02-12T15:30:00Z",
  "tool_version": "0.2.0",
  "page_count": 29,
  "pages": [
    {
      "slug": "1-overview",
      "title": "1 Overview",
      "depth": 0,
      "content": "...markdown content..."
    }
  ]
}

Useful for feeding individual pages into separate context windows, building retrieval indexes, etc.

How It Works

dw2md is a minimal MCP client that talks to DeepWiki's public JSON-RPC endpoint (https://mcp.deepwiki.com/mcp). No API key, no auth, no browser automation.

1. Initialize — MCP handshake with the server
2. Fetch structure — read_wiki_structure returns the table of contents
3. Fetch content — read_wiki_contents returns all pages in one response
4. Match & compile — split content by page markers, match to structure, assemble output

Failed requests are retried 3 times with exponential backoff (1s, 2s, 4s).

Contributing

Contributions welcome! Please ensure:

• All tests pass: cargo test
• Code is formatted: cargo fmt
• Clippy is happy: cargo clippy

License

MIT