website-to-api

website-to-api

A Claude Code plugin for reverse-engineering website internal APIs using Chrome browser automation.

What This Does

Most websites don't have public APIs, but they all have internal ones. This plugin provides:

1. A meta-skill (website-to-api) that teaches Claude how to discover and wrap any website's internal API
2. Site-specific skills (like substack) that implement the pattern for known sites
3. Templates for quickly adding new sites

How It Works

The pattern is:

1. Discover — Navigate to a website using the Claude in Chrome extension, inspect network requests, and find internal API endpoints
2. Authenticate — Extract httpOnly session cookies from the browser to use in scripts
3. Script — Build a CLI that calls the discovered endpoints with the extracted cookies
4. Adapt — When APIs change (they're internal and undocumented), re-discover and update

Installation

Add this plugin to your Claude Code project:

git submodule add https://github.com/hamelsmu/website-to-api.git

Available Skills

Skill	Description
website-to-api:website-to-api	Meta-skill: how to discover and wrap any website's internal API
website-to-api:substack	Retrieve posts from any Substack newsletter (list, search, full text)

Usage

Discovering a new site's API

Ask Claude:

"Can you figure out the API for [website]? I want to be able to list and download content from it."

Claude will use the website-to-api meta-skill to:

• Navigate to the site via Chrome
• Inspect network requests to find API endpoints
• Test authentication
• Build a script to call those endpoints

Using the Substack skill

# Set your auth cookie (extract from Chrome DevTools > Application > Cookies > substack.com)
export SUBSTACK_SID="your-cookie-value"

# List recent posts
uv run skills/substack/scripts/substack.py list-posts https://www.lennysnewsletter.com --limit 20

# Download a post as Markdown
uv run skills/substack/scripts/substack.py get-text https://www.lennysnewsletter.com some-post-slug

Adding a New Site

1. Copy the template: cp -r templates/site-skill-template skills/your-site
2. Use the website-to-api meta-skill to discover the site's API
3. Fill in the SKILL.md and script with the discovered endpoints
4. Submit a PR

Prerequisites

• Claude Code CLI
• Claude in Chrome extension (for API discovery and auth verification)
• Python 3.9+ and uv (scripts use PEP 723 inline metadata)

Design Principles

• Browser-first auth: Use the Chrome extension to handle authentication. The browser already has the user's session — don't ask humans for what the browser can provide.
• Resilient to change: Internal APIs break. Every site skill includes recovery instructions and references the meta-skill for re-discovery.
• No secrets in code: Auth cookies are always passed via environment variables, never stored in skill files or scripts.
• Self-contained scripts: Each script uses PEP 723 inline metadata so it runs with just uv run — no install step.