GitHub Docs Extractor

• Why?
• Features
• Output Structure & Naming Rules
  • 1. One-level directory structure
  • 2. Filenames encode the full relative path
  • 3. Output directory names are included in filenames
  • 4. Duplicate segments are avoided
• Installation
• Usage
  • Options
  • Examples
• Avoiding API Rate Limits
• Best Practices for Claude & AI Projects
• License

A powerful and simple CLI tool to download .md and .mdx documentation files from any public or private GitHub repository. It prepares the files for optimal use in Claude Projects, AI assistants, and Retrieval-Augmented Generation (RAG) systems by extracting documentation into a clean, predictable, and AI-friendly structure.

Why?

When using AI models like Claude or building RAG systems, the structure and naming of your documents matter as much as the content itself. Flattening folder structures while preserving semantic context in filenames is critical for high-quality document retrieval.

Claude performs best when:

• Documents are logically grouped
• Files are small, focused, and descriptively named
• Folder depth is shallow and intentional

github-docs-extractor automates the extraction of GitHub documentation so it can be uploaded directly into a Claude Project or AI knowledge base with minimal manual cleanup.

Features

• Targeted Extraction: Pull docs from one or more specific folders in a repository.
• Claude-Ready Output: Produces a structure aligned with Claude Project best practices.
• Smart Filenaming: Generates descriptive, human-readable filenames.
• Deterministic Filenaming: Filenames always encode the original documentation path to preserve context.
• Category-Friendly: Designed to work naturally with one-level subfolder organization.
• ZIP Archiving: Optionally create a single .zip file for easy upload to AI platforms.
• User-Friendly CLI: Interactive spinners and colored output.
• Alias Support: Use the shorter gde command.
• Works with Private Repositories: Supports GitHub Personal Access Tokens (PAT). See Avoiding API Rate Limits.

Output Structure & Naming Rules

This tool intentionally flattens documentation structures to make them optimal for AI ingestion (Claude Projects, RAG systems, etc.).

The output always follows deterministic and predictable rules designed to preserve semantic context while avoiding deep folder hierarchies.

1. One-level directory structure

All extracted files are written into a single output directory level.

The name of this directory is derived from the documentation path provided using the --paths option.

| --paths value | Output directory | |-----------------|------------------| | docs | output/ | | packages/react/docs | output/react-docs/ | | examples/with-mdx | output/examples-with-mdx/ | | docs/code | output/code/ |

Documentation container names such as docs, doc, and documentation are treated as neutral and are ignored when computing the output directory name.

2. Filenames encode the full relative path

To preserve context after flattening, filenames always encode the full relative path under the documentation root.

This ensures that no semantic information is lost, even when multiple files share the same base name.

Example input structure:

docs/
└── code/
    ├── api.md
    └── api/
        └── api.md

Extracted output:

output/code/
├── code-api.md
└── code-api-api.md

Each folder name under the documentation path becomes part of the filename, joined using hyphens (-).

3. Output directory names are included in filenames

When an output directory exists, its name is always included as the first segment of the filename.

Example:

--paths docs/code

docs/code/get-started.md will be saved at output/code/code-get-started.md

This guarantees that filenames remain self-descriptive and retain their categorical context even when uploaded into flat or semi-flat systems.

4. Duplicate segments are avoided

If a filename already starts with the expected prefix, it will not be duplicated.

This prevents redundant or unreadable filenames while keeping the naming rules fully deterministic.

Installation

Install globally via npm:

npm install -g github-docs-extractor

Or run directly using npx:

npx github-docs-extractor --repo <url> ...

Usage

The command is straightforward. You must provide a repository URL. Other options are available to customize the extraction.

gde --repo <repository_url> [options]

You can also use the full name github-docs-extractor.

Options

| Option | Alias | Description | Default | | ------ | ----- | ----------- | ------- | | --repo <url> | -r | Required. The full URL of the GitHub repository. | | | --paths <paths...> | -p | One or more space-separated paths to the documentation folders in the repo. | docs | | --out <dir> | -o | The destination directory for the downloaded files. | ./output | | --zip | | If set, creates a zip archive of the output folder. | false | | --version | -V | Display the version menu. | | | --help | -h | Display the help menu. | |

Examples

1. Basic Usage (extracting the default docs folder)

gde --repo "https://github.com/adrienv1520/claude-master"

This will download all markdown files from the docs/ folder of the Claude Master documentation repository into a local ./output directory with only one subfolder level.

2. Specifying a custom path and output directory

gde -r "https://github.com/facebook/react" -p "packages/react-dom/docs" -o "./react-dom-docs"

This extracts only the docs related to react-dom and saves them in ./react-dom-docs with only one subfolder level if there was any.

3. Using a multiple paths

gde \
  --repo "https://github.com/vercel/next.js" \
  --paths docs examples/with-mdx \
  --out "./nextjs-docs"

This will:

• Fetch files from both docs/ and examples/with-mdx/.
• Save them in ./nextjs-docs with only one subfolder level if there was any.

4. Creating a ZIP archive

gde \
  --repo "https://github.com/mdn/content" \
  --paths "files/en-us/web/javascript" \
  --out "./mdn-javascript" \
  --zip

This command downloads the entire MDN JavaScript documentation, saves it to ./mdn-javascript, and then creates a mdn-javascript.zip file in the same parent directory, ready for upload.

Avoiding API Rate Limits

The GitHub API has different rate limits for requests:

• Unauthenticated: ~60 requests per hour (from your IP address).
• Authenticated: ~5,000 requests per hour (using a Personal Access Token).

This tool can make many requests for large repositories. To avoid hitting the lower limit, it is highly recommended to use a Personal Access Token (PAT) for all uses, including on public repositories. Using a PAT is also required to access private repositories.

How to Use a Personal Access Token:

1. Create a new Personal Access Token on GitHub (the "classic" version is fine).

   • For accessing public repos, grant the repo/public_repo scope.
   • For accessing private repos, grant the repo scope.

2. Set it as an environment variable named GITHUB_TOKEN before running the command. The CLI will automatically detect and use it.

# For macOS / Linux
export GITHUB_TOKEN="your_personal_access_token_here"

# For Windows (Command Prompt)
set GITHUB_TOKEN="your_personal_access_token_here"

# For Windows (PowerShell)
$env:GITHUB_TOKEN="your_personal_access_token_here"

Best Practices for Claude & AI Projects

This repository is aligned with Claude's own recommendations you can find here: Claude Master.

By using github-docs-extractor, you are aligning your documentation with Claude’s native strengths, resulting in better search, better context selection, and better answers.

License

MIT.