What Loreto Does

Loreto turns any content source—YouTube videos, articles, PDFs, or images—into structured, ranked Claude Code skill packages in a single generate_skills tool call.

Each package includes:

A complete SKILL.md
README.md
Reference files
A runnable test script

All files are ready to save directly to .claude/skills/.

Instead of watching a 45-minute conference talk and manually distilling the principles into a skill, you can paste the URL into Claude Code and get back multiple ranked skills. This includes abstract principles buried in case studies and business narratives that might otherwise be missed.

Tools

generate_skills

Extracts 1–5 ranked skill packages from a source.

Capabilities

Supports:
- YouTube (up to 60 minutes)
- Web articles (up to 50k words)
- PDFs (up to 100 pages)
- Images/diagrams (up to 20 MB)
Automatically detects source type from the URL
Ranks each skill by how concrete and code-executable it is
Fully scaffolds up to 3 skills per call
Queues remaining themes with exact names for follow-up requests

Configuration

Test language: Python / TypeScript / JavaScript
Mermaid diagrams: on/off
Optional context hint

get_quota

Check your API usage and remaining calls for the current billing period before starting a large extraction session.

What You Get Back

Each skill package contains:

SKILL.md — Core principles, failure modes, implementation steps, and architectural patterns
README.md — Overview and usage context
Reference files — Supporting patterns, data structures, and domain-specific content
Test script — Runnable validation in your chosen language

The response also includes:

A theme plan showing:
- All detected themes
- Which were processed
- Which are queued
Token usage broken down by pipeline stage

How It Works

A single source can teach multiple lessons.

Loreto’s pipeline identifies every generalizable principle in the content—not just the obvious headline topic—and ranks them by how directly applicable they are to code.

For example, a single technical talk might yield:

Database migration strategy (rank 1)
On-call culture design (rank 2)
Stakeholder communication patterns (rank 3)

Pipeline Characteristics

Runs within an 8-minute timeout
Applies quality gates to filter low-signal themes
Returns partial results if some themes pass and others don’t

Setup

Install:

uvx loreto-mcp

Add to ~/.claude/mcp.json (user-scoped) or .mcp.json (project-scoped):
{
  "mcpServers": {
    "loreto": {
      "command": "uvx",
      "args": ["loreto-mcp"],
      "env": {
        "LORETO_API_KEY": "lor_..."
      }
    }
  }
}

Get your API key at https://loreto.io

> Extract skills from https://www.youtube.com/watch?v=JYcidOS9ozU and save them to .claude/skills/

> Turn this architecture diagram into reusable skills: https://...

> Check my Loreto quota, then extract skills from these three articles.

Claude handles the tool calls, receives the full file contents, and writes them directly into your project.