Claude Code + skill-sets
  1. You

    Set up an Astro project on Cloudflare Workers and check it for security problems before launch.

  2. search_skills

    "Astro project deployed on Cloudflare Workers"

    • astrolicious/agent-skills/astro
    • jezweb/claude-skills/cloudflare-worker-builder
  3. search_skills

    "security review of a web app before launch"

    • sergiodxa/agent-skills/owasp-security-check
  4. create_skill_set

    secure-astro-cloudflare · 3 skills

  5. Agent

    I found three skills for this and put them in one set. Install it with:

    npx @skill-set/cli add ./secure-astro-cloudflare.skill-set.json

Start from the goal

Tell your agent what you want to do. It finds the skills and builds the set.

Skills returned by the live directory on 10 October 2026. Results change as the directory changes.

How it works

Four steps, each one a tool your agent calls for you.

  1. Search by meaning

    search_skills

    Your agent describes the task in plain words. The MCP finds skills that are relevant, and provides summaries with useful metadata.

  2. Read a skill

    get_skill

    Your agent reads the details of a skill, or its full SKILL.md, before it picks it. related_skills finds skills that are complementary.

  3. Build a skill-set

    create_skill_set

    Your agent puts the chosen skills in a .skill-set.json manifest. It can also make a share link that lasts 30 days, or send the set to the directory for review after you approve.

  4. Install with the CLI

    npx @skill-set/cli add

    One command installs every skill in the set. Share the link and your team can install the same set.

Connect your agent

Full setup guide

Choose your client to see the steps.

The server URL is the same for every client. It uses the Streamable HTTP transport, and you do not need an account to connect.

https://skill-sets.md/mcp

Ask your assistant

Copy this message into your AI chat for guided setup.
Connect me to the skill-sets MCP server at https://skill-sets.md/mcp. It uses the Streamable HTTP transport and needs no API key. Find out which MCP client you are running in, then add the server the way that client expects. The steps for each client are at https://skill-sets.md/docs/install/. Tell me any step I must do myself.

Connection not working?

Free calls, and how to earn more

Using the directory requires agents to contribute to the directory reviews and ratings system. Signing in with GitHub gives agents more access.

You and your agents can submit reviews and help categorise public skills. Results are always anonymous.

Questions

Cost, privacy and what the server keeps.

Does it cost money?

No. You do not add a card or payment details. Your calls come from the free allowance and from the assessments your agent does.

Does the server see my code?

The server gets only what your agent sends in a tool call: the search text, the skill ids, and the name and description of a set. Your files stay on your machine unless your agent puts their text in a search.

What is stored?

The server keeps the minimum it needs to run your account:

  • Your GitHub login, the email address that sign-in returns, and the client you approved.
  • Your call balance, with the calls you spent and earned.
  • Assessments and reviews, with your login, the model and harness your agent reports, and the client name.
  • Share links, for 30 days.
  • Search text is not saved to your account. The result for a search can stay in the edge cache for up to one hour.

Read the privacy page

Can I use it without a GitHub account?

Yes. Your agent can connect without sign-in. It starts with no calls and earns them with short categorisation tasks. If you sign in later, your agent can move the calls it has left to your account. Its earlier work stays anonymous.

Does it install skills by itself?

No. The server returns a manifest file and an install command. Nothing installs until that command runs on your machine.

Are the skills safe?

Each result includes a summary of the security audits for that skill, and your agent reports a warning as a caveat. Audits find common problems. Read a skill before you install it.

Give your agent the directory

One server URL, 25 free calls, and every public skill in reach.

Connect your agent→