Convert Docs to Claude Skills

Skill Seekers is a CLI and MCP server that acts as a universal preprocessing layer between raw documentation and the AI systems that need to consume it. It ingests 18 source types - documentation websites, GitHub repositories, local codebases, PDFs, Word documents, EPUB e-books, Jupyter notebooks, OpenAPI/Swagger specs, PowerPoint presentations, AsciiDoc, local HTML, RSS/Atom feeds, man pages, Confluence wikis, Notion pages, Slack/Discord chat exports, and videos (YouTube, Vimeo, or local files) - and converts them into a single structured knowledge asset. The pipeline runs in four stages: scrape, categorize, AI-enhance (producing a 500+ line SKILL.md with examples and patterns), and package. A one-command install workflow (skill-seekers install --config react) chains fetch, scrape, mandatory AI enhancement, packaging, and optional upload to Claude in 20-45 minutes. An AI-driven scan command can also read a project's manifests, README, Dockerfile/CI, and sampled source imports to auto-generate one config per detected framework.

Use it when you need to prepare the same documentation once and export it to many AI targets, instead of manually re-formatting docs for each platform. Typical jobs: building a Claude AI Skill, populating a LangChain or LlamaIndex RAG pipeline, upserting into a vector database, or generating .cursorrules/.windsurfrules context files for an AI coding assistant. The project claims this collapses days of manual data prep into 15-45 minutes, backed by 3,700+ tests.

Input: any of the 18 supported source types, via skill-seekers create <source> (a URL, a GitHub owner/repo, a local path, or a file). Output: a packaged skill directory that skill-seekers package output/<name> --target <platform> converts into a platform-specific artifact - a Claude Skill ZIP+YAML, a Gemini tar.gz, an OpenAI ZIP with vector store, LangChain Document JSON, LlamaIndex TextNode JSON, Haystack Documents, or Markdown for Pinecone/Chroma/FAISS/Qdrant upsert. GitHub sources get an additional three-stream analysis (Code via AST parsing, Docs, and Insights from issues/stars/forks) with conflict detection between documented and actual APIs.

Twelve LLM platforms: Claude AI, Google Gemini, OpenAI ChatGPT, MiniMax AI, Generic Markdown, OpenCode, Kimi (Moonshot AI), DeepSeek AI, Qwen, OpenRouter, Together AI, and Fireworks AI. RAG frameworks: LangChain, LlamaIndex, Haystack, and vector databases Pinecone, Chroma, FAISS, Qdrant, Weaviate. AI coding assistants: Cursor, Windsurf, Cline, and Continue.dev, plus one-command install to 19 AI coding agents including Claude Code, VS Code/Copilot, Amp, Goose, Roo Code, Aider, Bolt, Kilo Code, and IBM Bob. The AI-enhancement step does not require an Anthropic API key: it can run against an existing Claude Pro/Max, ChatGPT Plus, or GitHub Copilot subscription via --agent claude/codex/copilot/opencode/kimi, or against any OpenAI-compatible endpoint (OpenRouter, Groq, Cerebras, Mistral, NVIDIA NIM) via environment variables. It also ships an MCP server exposing 40 tools (scraping, packaging, uploading, vector-DB export, cloud storage) for Claude Code, Cursor, Windsurf, and VS Code + Cline.

skill-seekers create https://docs.react.dev/
skill-seekers package output/react --target claude

AI Skill builders working with Claude, Gemini, or OpenAI who need production-grade SKILL.md files; RAG engineers who need pre-chunked, metadata-rich documents for LangChain, LlamaIndex, or a vector database; and teams standardizing AI coding assistant context (Cursor, Windsurf, Cline, Continue.dev) across a codebase without re-scraping for each tool.