Data study · v1, October 2026

Agent skill and AI tool directories: a dated comparison

All editions of this study

The short answer

This is a dated comparison of agent skill and AI tool directories, not a ranking. Sizes are each site's own claim, quoted as the site states it and read on , unless a cell is labelled as our sitemap count; we did not audit catalogue contents or remove duplicates. In our Claude tests, all 18 searches about little-known skills used web search, and 13 of the 18 answers linked a third-party skill directory in the answer body. In our Claude tests, There's An AI For That appeared in 27 of 27 search runs for the three broad directory questions, and Futurepedia in 27 of 27 as well.

Key takeaways

  1. In our Claude tests, skills.lc was linked in 6 answers, MCP Market in 4 answers and LobeHub in 2 answers.
  2. In our Claude tests, all 18 searches about little-known skills used web search, and 13 of the 18 answers linked a third-party skill directory in the answer body.
  3. In our Claude tests, There's An AI For That appeared in 27 of 27 search runs for the three broad directory questions, and Futurepedia in 27 of 27 as well.
  4. Every site figure in the tables was read on ; sizes are the sites' own claims unless we label a sitemap count.
  5. Some sites stated different sizes in different places on the same day: LobeHub's index said 334,168 skills and its llms.txt said 361328+.

This is a dated comparison, not a ranking. CookiesForAI is also a directory of agent skills, MCP servers and AI coding tools, so we leave it out of the tables. Sizes below are each site's own claim, quoted as the site states it, unless a cell is labelled as our sitemap count.

In our Claude tests, all 18 searches about little-known skills used web search, and 13 of the 18 answers linked a third-party skill directory in the answer body. In our Claude tests, skills.lc was linked in 6 answers, MCP Market in 4 answers and LobeHub in 2 answers.

For broader directory questions, the pattern was different. In our Claude tests, There's An AI For That appeared in 27 of 27 search runs for the three broad directory questions, and Futurepedia in 27 of 27 as well. These results describe those Claude tests on that date, not current visibility or quality.

Which tool directories appeared in the broad directory questions?

  1. There's An AI For That27 of 27
  2. Futurepedia27 of 27

Search runs that named each directory, for the three broad directory questions in our Claude tests.

Source: CookiesForAI, v1, October 2026, 27 runs

Reuse this chart

Free to reuse with credit (CC BY 4.0). Download SVG

Embed code
<figure><img src="https://cookiesforai.app/charts/agent-skill-and-ai-tool-directories-compared-october-2026--broad-directory-runs.svg" alt="Which tool directories appeared in the broad directory questions: There's An AI For That 27 of 27, Futurepedia 27 of 27."><figcaption>Source: CookiesForAI, Which tool directories appeared in the broad directory questions?, https://cookiesforai.app/research/agent-skill-and-ai-tool-directories-compared-october-2026</figcaption></figure>

What do the agent skill directories offer?

Every site figure in this table was read on .

Directory Size as the site states it What its pages offer Listing route and paid options Agent-facing access
LobeHub The index said "Browse 334,168 Skills built agent-first". On the same day, llms.txt said "(361328+ skills)". The example page includes author, date, category, license, usage counts, a summary, SKILL.md and version history. We could not verify a publishing route that day. llms.txt links an OpenAPI file, a WebMCP endpoint and a pricing document; llms-full.txt is also present.
MCP Market The home page said "49,125 Servers". The submit page claimed "35K+" MCP servers and "200K+" agent skills. The example server page includes author, stars, categories, an official label, description, related servers, features, use cases and FAQ headings. The submit page offers a free queue and a paid fast-track. llms.txt links Markdown hub pages. Its robots.txt disallows ClaudeBot and several other named crawlers.
agentmods The site said "614,248 mods indexed" and "All 425,363 Skills". The example page includes token cost, scan grade, provenance, license, stars, an install command and a SKILL.md excerpt. The home page links to a page for submitting a repository. llms.txt describes the catalogue, an MCP endpoint and a CLI.
skills.rest The home page said "754,000+ skills indexed". On the same day, llms.txt said "113,000+ vetted agent skills". The example repository page includes a community label, an install command and descriptions for its included skills. Its contribute page accepts a GitHub repository URL and describes screening and indexing. llms.txt is present.
Skills Directory The home page and llms.txt said "649,882 indexed skills". The example page includes a grade, stars, scan date, compatibility, an install command, SKILL.md and a badge URL. The home page links to submission; the site also offers paid sponsor slots and a paid membership. llms.txt says a public API is available.
skills.sh The leaderboard showed "All Time (1,294,300)". The example page includes an install command, SKILL.md, installs, stars, first-seen date and security audit results. Its docs say the leaderboard is based on anonymous install telemetry from its CLI. The item page provides the CLI command. Its llms.txt page did not load for our reader.
SkillsLLM The index said "All Skills ( 5760 found)". On the same day, llms.txt said "1,600+ open-source AI agent skills". The example page includes the author, verification label, stars and description. A submit page exists, though the page we read did not show its form. Its sponsorship policy says self-service featured placements are currently unavailable. llms.txt is present.
AgenticSkills The site said "193+ skills indexed", "196 MCP servers" and "9 platforms tracked". The example page includes editorial rank, author, supported platform, stars, recency and CLI or manual installation. Its submission page says listing is free and selective, and that featured placement is available. Its llms.txt page did not load for our reader.
skills.lc Its page did not expose a size to our reader. The item page did not expose enough content to our reader to verify its fields. We did not check a listing route. Its llms.txt address returned the site's HTML shell.

What do the AI tool directories offer?

Every site figure in this table was read on . Where a page did not load for our reader, we do not repeat an older number.

Directory Size as the site states it, or our count What we could read Listing route and paid options Agent-facing access
There's An AI For That Its home page did not load for our reader. Its item and submit pages did not load for our reader. We could not verify its listing options. Its robots.txt disallows ClaudeBot and several other named crawlers.
Futurepedia Our sitemap count was 1,332 tool entries. The example page includes a rating, user reviews, verification, categories, an overview and alternatives. We did not find a submission page that day. Its llms.txt page did not load for our reader.
Toolify Its home page did not load for our reader. Its item and submit pages did not load for our reader. We could not verify its listing options. Its robots.txt was readable; its llms.txt page did not load for our reader.
AITopTools The home page said "11,000+ AI tools". Our sitemap count found 11,815 entries, including 11,200 tool URLs. The example page includes verification, a rating, category and an overview. Its submission page builds a listing draft from a tool URL. Its llms.txt page did not load for our reader.
ToolDirectory.ai Its home page did not load for our reader. Its item, pricing and submit pages did not load for our reader. We could not verify its listing options. llms.txt is present and mentions an MCP server.
AIChief Its llms.txt said "organized in 649+ categories with 12,196+ AI tools". Its item and submit pages did not load for our reader. We could not verify its listing options. llms.txt and robots.txt were readable.

What is useful to check across the directories?

The pages vary, but a reader choosing where to browse or publish can check the same practical details: whether an item has a stable page, whether the source and author are visible, whether installation is clear, how screening is described, and whether machine-readable access exists. A large self-reported catalogue and a detailed item page answer different questions, so we show them separately.

Paid options are also not equivalent. The readable first-party pages describe paid fast-tracks, paid sponsor slots, paid memberships or featured placement. We link those sites' own pages rather than quoting third-party prices.

What are the limits?

  • This is a snapshot. Site figures were read on and can change.
  • Sizes are the sites' own claims unless we explicitly label a sitemap count. We did not independently audit catalogue contents or remove duplicates.
  • A page that did not load for our reader may work in a browser. We treat it as unreadable for this comparison, not as evidence that the site or feature is unavailable.
  • The Claude findings come only from our Claude tests. They do not describe other assistants, later runs or a causal effect of page features.
  • We describe only details visible in the first-party pages we read. If a form, API or other feature was not visible there, we do not infer it from an older page or another publisher.

Changes on : re-read every site figure; removed counts from our six-engine notes and listing fees from a third-party article; updated changed directory figures; replaced figures we could not verify with a plain note that the page did not load for our reader; and limited test results to the documented Claude counts from .

Data and how to cite

Download the chart data: CSV, JSON. Free to reuse with credit (CC BY 4.0).

How to cite
CookiesForAI, "Agent skill and AI tool directories: a dated comparison", v1, October 2026, https://cookiesforai.app/research/agent-skill-and-ai-tool-directories-compared-october-2026

CookiesForAI lists agent skills, MCP servers and apps, so our own pages can appear in these answers; when they do, we count them like any other site.

Questions people ask

What do the directory sizes in this comparison mean?

They are each site's own claim, quoted as the site states it and read on , except where we explicitly say that we counted sitemap entries.

What did Claude cite in the agent skill tests?

In our Claude tests, skills.lc was linked in 6 answers, MCP Market in 4 answers and LobeHub in 2 answers.

Sources

Changes to this page

  • : page back after we took it down on to check every figure again; rewritten with every site figure re-read on , Claude counts from the test transcripts, and the six-engine counts and third-party fees removed.
  • : short answer, takeaways and charts added; section headings now ask questions.