Common questions about AI skills, SKILL.md, MCP servers, and how to use SkillsLLM.
An AI skills marketplace is a curated directory of reusable capabilities — agents, MCP servers, CLI tools, IDE extensions, and other instruction packages — that extend AI coding assistants like Claude Code, Codex CLI, and ChatGPT. Marketplaces aggregate, categorize, and (in SkillsLLM's case) security-scan open-source skills so developers can discover and install them without combing through GitHub.
SKILL.md is a markdown file at the root of a GitHub repository that describes an AI skill: its purpose, capabilities, installation instructions, and supported AI assistants. It is Anthropic's emerging standard for packaging modular AI capabilities. When Claude Code or another AI assistant loads a skill, it reads SKILL.md to understand what tools and instructions to apply.
Model Context Protocol (MCP) is an open standard from Anthropic that connects AI assistants to external tools, databases, APIs, and file systems. An MCP server exposes tools (functions the AI can call) and resources (data the AI can read) over a defined protocol. SkillsLLM indexes 200+ MCP servers across categories like databases, productivity, search, and dev tools.
Most Claude Code skills are installed by cloning the GitHub repository into your Claude Code workspace's skills directory: `git clone https://github.com/owner/skill-name`. Some skills require additional setup (npm install, environment variables, MCP server configuration). Each skill page on SkillsLLM links to the original GitHub repo where exact installation instructions live in the README.md or SKILL.md.
Browse the MCP Servers category on SkillsLLM at /category/mcp-servers. SkillsLLM filters by minimum 100 GitHub stars to keep low-effort projects out of results. You can sort by stars, recency, or community votes, and use the search box to find servers for specific use cases (Postgres, GitHub, Linear, etc.).
The most popular Claude Code skills on SkillsLLM by stars and community votes include Caveman (token compression), Context7 (up-to-date docs), Cursor Rules (workspace conventions), and various MCP servers for databases and productivity tools. Browse trending and new-this-week sections on the homepage, or filter by category to find skills matching your workflow.
Create a SKILL.md file at the root of your GitHub repository describing the skill name, purpose, capabilities, installation, and supported AI assistants. Submit your repository at https://skillsllm.com/submit (GitHub OAuth required). The scraper validates the repo, fetches metadata, runs a security scan, and adds it to the catalog within 24 hours.
Yes. Browsing, searching, voting, bookmarking, commenting, and submitting skills are all free. No signup is required to browse. GitHub OAuth is required for community features (votes, bookmarks, comments, submissions). SkillsLLM does sell separate paid courses on agentic AI, but the skills marketplace itself is free.
Every skill is scanned daily by a CI workflow that clones the repository, runs Semgrep (static code analysis for known vulnerability patterns), npm audit (for JavaScript dependency CVEs), and pip-audit (for Python dependency CVEs). Results are summarized as PASS, WARNING, or FAIL on the skill page, with the full report visible to anyone. No skill is hidden — but flagged skills are clearly labeled.
SKILL.md is a metadata file that describes any AI skill — including non-MCP ones like CLI tools, agents, and code-gen helpers. MCP (Model Context Protocol) is a runtime protocol for connecting AI assistants to external tools and data. Many MCP server repos include a SKILL.md describing the server, but SKILL.md and MCP are independent: a skill can have SKILL.md without being an MCP server, and an MCP server can exist without SKILL.md.
Yes. The active niche is data-driven sports betting and prediction markets rather than casino games: skills that pull live odds, compare books, flag value (positive-EV) bets, size stakes with the Kelly criterion, and read prediction markets like Polymarket and Kalshi. SkillsLLM curates them on the Sports Betting & Prediction Markets hub at /hub/sports-betting, which uses a lower star threshold than the main 100-star catalog because the niche is young. These are developer tools for analysis and automation — not betting advice, and not casino-game bots.
Not reliably. Most 'AI casino' projects on GitHub are entertainment demos that simulate blackjack, roulette, or slots rather than help a player win, and card-counting bots violate every real casino's terms of service. In 2026 head-to-head tests, LLMs apply foundational poker theory competently but do not optimize through simulation like dedicated solvers do. The genuinely useful tooling is in sports betting and prediction markets, where the work is a public-data problem — see /hub/sports-betting.