SkillsMP vs SkillzBench: Which Agent Skills Platform Should You Trust?
SkillsMP is a free discovery index of over 3.6 million SKILL.md agent skills, searchable by keyword, occupation, creator, and category, with a free tier that allows 1 export task per day and 100 API/MCP searches per day. SkillzBench is also free, but it goes further than discovery: it verifies, security-scans, and benchmarks every skill in a sandbox, publishes a measured Lift score, and gives you a free MCP connector with per-agent loadouts. If you just want to browse what exists, SkillsMP works; if you need to know which skills are safe and actually improve your agent, SkillzBench is the stronger choice.
TL;DR
- SkillsMP reports 3,601,307 collected SKILL.md files, making it the largest raw discovery index of the two
- SkillsMP's own terms say it does not endorse or verify the quality, safety, or functionality of any indexed skill
- SkillzBench is fully free: directory, MCP connector, sandbox benchmarks, and change alerts, funded by blog advertising rather than user fees
- SkillzBench pins every skill version to a commit SHA and runs security, license, and quality checks before listing
- SkillzBench benchmarks each skill in a sandbox with and without the skill and publishes the Lift in percentage points
- SkillsMP's free tier allows 1 export task per day (up to 500 rows) and 100 API/MCP searches per day
- Published research found roughly 26% of agent skills carry at least one vulnerability, which is why verification matters
Head to head
| Dimension | SkillsMP | SkillzBench |
|---|---|---|
| Pricing | Free; 1 export task/day (500 rows) and 100 API/MCP searches/day | Everything free: directory, MCP connector, benchmarks, alerts |
| Catalog scale | 3,601,307 collected SKILL.md files across 867 SOC occupations 🏆 | Hundreds of deduplicated, verified skills from GitHub |
| Trust and safety | No verification; terms disclaim quality, safety, and functionality | Commit-SHA pinning, security scans, license checks, 0-100 trust score 🏆 |
| Benchmarking | None; publication is explicitly not a ranking or certification | Sandbox test with and without each skill; Lift published in percentage points 🏆 |
| Agent integration | API/MCP search access, 100 searches/day on the free tier | Free MCP server plus per-agent loadouts controlling what each agent sees 🏆 |
| Discovery and browsing | Keyword, occupation, creator, category, repository; curated collections 🏆 | Directory organized around verified skills with trust scores and Lift |
| Ongoing monitoring | Cards can lag behind source repositories | Tracks upstream changes, re-scans, re-benchmarks, and notifies 🏆 |
The contenders
SkillsMP (Agent Skills Marketplace) is an independent community project that indexes public SKILL.md files from GitHub and makes them searchable by keyword, occupation, creator, category, and repository. It is not affiliated with Anthropic or OpenAI. The homepage reports 3,601,307 collected SKILL.md files, organized across 23 occupation groups and 867 detailed occupations based on the U.S. Standard Occupational Classification, with category counts like Tools (831,744), Business (656,489), and Development (470,979). It surfaces curated collections from known creators, such as Matt Pocock's Skills For Real Engineers (38 skills) and Corey Haines' Marketing Skills (50 skills covering copywriting, SEO, and campaigns). Each skill card links to its public GitHub source so you can inspect the SKILL.md and companion files before installing. SkillsMP positions itself as a discovery map and reference library: it indexes and organizes but does not certify, install, or safety-review skills, and its terms of service put the responsibility for reviewing code squarely on you.
pricing 0
- + Massive catalog: over 3.6 million SKILL.md files, so you can almost always find references for a task
- + Completely free to browse and search, with a documented free tier for exports and API/MCP access
- + Strong browsing structure: occupations, categories, creators, and curated collections make discovery easy
- + Every card links to the public GitHub source with star counts and last-activity dates for manual vetting
- + Review-first install prompt is selected by default, nudging you to read files before installing
- − Its own terms state it does not endorse or verify the quality, safety, or functionality of any skill
- − No sandbox benchmarking, so you get no measured evidence a skill actually improves your agent
- − No trust score, license check, or security scan; auditing code before install is entirely your job
- − Individual cards can lag behind their source repositories, so indexed data may be stale
- − Free tier caps exports at 1 task per day and API/MCP searches at 100 per day
SkillzBench is AIQ Labs' free, independent directory of AI agent skills plus a free MCP tool that connects any agent to the skills you choose. It was built around a simple problem: agent skills (SKILL.md instruction packages for Claude Code, Codex, Cursor, Gemini CLI, and chat-app agents) are everywhere, but published research found roughly 26% of them carry at least one vulnerability, and nobody could tell which ones actually help. SkillzBench continuously crawls and dedupes skills from GitHub, then pins every version to a commit SHA, validates it, checks its license, runs a security scan, and judges quality. Every skill is benchmarked in a sandbox with the same model, with and without the skill, and the resulting Lift is published in percentage points, always shown separately from ratings. A 0-100 trust score combines security, quality, license, maintenance, and community signals, and money never touches scores. You equip agents through one free MCP server plus per-agent loadouts, toggling skills and tools so each agent only sees what you chose. SkillzBench tracks upstream changes, re-scans, re-benchmarks, and notifies you. Everything is free; revenue comes from advertising around blog articles, never from users.
pricing 0
- + Everything is free, including the MCP connector, benchmarks, and alerts; no paid tiers at all
- + Addresses the ~26% vulnerability rate in published research with security scans and license checks on every skill
- + Lift benchmarking gives you measured evidence of whether a skill actually helps, not just a description
- + Per-agent loadouts let you control exactly what each agent can see, which matters for safety
- + Trust scores are transparent and ad revenue never influences scores or listings
- + Upstream monitoring means you hear about changes to skills you rely on
- − The verified catalog is much smaller than SkillsMP's raw index of millions of files
- − Focused on skills quality and safety, so it is not a general marketplace for browsing by occupation or creative field
- − Revenue comes from blog advertising, so you will see ads around editorial content
Verdict
Winner: SkillzBench
Both platforms are free, so the decision comes down to trust. SkillsMP indexes millions of skills but explicitly does not verify quality or safety, while SkillzBench pins versions, scans for vulnerabilities, and benchmarks every skill with a published Lift figure.
| Option | Overall | Breakdown |
|---|---|---|
| SkillsMP | 6.5 | : 9: 10: 3: 8: 2: 6 |
| SkillzBench | 8.5 | : 10: 9: 9: 6: 9: 8 |
Pricing compared
Trust and safety: the real differentiator
Scale versus diligence
Which should you pick?
- :
- :
- :
- :
- :
- :
FAQ
Are SkillsMP and SkillzBench both free?
Yes. SkillsMP's pricing page lists a single Free plan at US$0 with no paid billing cycle, though the free tier limits you to 1 export task per day (up to 500 rows) and 100 API/MCP searches per day. SkillzBench is also entirely free: the directory, the MCP connector, benchmarks, and alerts all cost nothing, because SkillzBench earns revenue from advertising around its blog articles rather than charging users, and ad money never influences scores or listings.
Does SkillsMP verify that skills are safe to install?
No. SkillsMP's own terms of service state that it does not endorse or verify the quality, safety, or functionality of any indexed skill, and that skills are provided as-is without warranties. The homepage similarly describes its listings as a discovery aid, not a ranking, certification, or safety review. You are expected to review the SKILL.md and companion files on GitHub yourself before installing anything.
What is Lift and why does it matter?
Lift is SkillzBench's benchmarking metric. Every skill is tested in a sandbox with the same model, once with the skill enabled and once without, and the difference in performance is published in percentage points. This gives you measured evidence of whether a skill actually helps your agent rather than just a well-written description. Lift is always shown separately from ratings, so a skill with a high trust score and a low Lift is easy to spot.
Why should I worry about skill vulnerabilities at all?
Because the risk is documented, not theoretical. Published research found that roughly 26% of agent skills carry at least one vulnerability. A skill is an instruction package that your agent will follow, and a malicious or poorly written one can direct the agent to run scripts, fetch URLs, or access sensitive files. That is why verification steps like commit-SHA pinning, security scanning, and license checking matter before anything reaches a production agent.
Can I use both platforms together?
Yes, and for many teams that is the sensible workflow. Use SkillsMP's massive index to discover what kinds of skills exist for your task and to study how experienced creators structure them. Then check the specific skills you are considering on SkillzBench, where trust scores, security scans, and published Lift results tell you whether a given skill is safe and actually improves performance. SkillzBench's free MCP connector then lets you equip the verified skills you chose to any agent.
Which platform is better for non-technical business owners?
SkillzBench. Its trust scores (0-100), published Lift benchmarks, and free MCP connector with per-agent loadouts are designed so you can make safe choices without reading code. SkillsMP assumes you will open the GitHub source, read the SKILL.md and companion files, and judge safety yourself, which its terms make explicit. If you have no developer available, SkillzBench's verification layer does that work for you.