Documentation · Install guide
Install in seconds.
Free skills: no key. MCP tools: need a free API key. Start with the one paste below — it needs no account at all.
Recommended · one paste
Using Claude Code, Codex or Cursor? Paste this into the chat
Install the loopskill skill from https://app.loopskill.io/skill Paste this into your agent's chat (Claude Code, Codex, Cursor…) and press Enter.
Your agent downloads the LoopSkill skill file (plain Markdown) and can then find and install free skills for you — no key, no config files. Works the same in Hermes, OpenClaw and any agent that can read a web page.
Signed in? Your library shows the same phrase with your API key already in it. You only need the key for MCP tools, saving to bundles, private content and fleets.
Other ways
You do not need these if the paste above worked. Pick one only if your setup needs it.
MCP server · needs a free API key
Connect your agent to the LoopSkill server
MCP is a standard way for an AI app to use outside tools. Cursor, Claude Desktop, Cline,
Continue.dev, Zed — every MCP-aware app connects the same way: you give it a URL and your key.
Your agent then gets loopskill_search,
loopskill_install,
and loopskill_recall
as tools it can call mid-conversation.
The MCP server requires an API key for all operations — including free-tier skill lookups. The key is free. If you want no key at all, use the one paste above, the skill file or the plain HTTP calls below.
Claude Desktop / Cursor / Cline
{
"mcpServers": {
"loopskill": {
"url": "https://app.loopskill.io/api/mcp/http/",
"headers": { "x-api-key": "rec_live_…" }
}
}
} Get your key from your library — free signup gives you a free-tier key immediately.
Your app can only start a local program, not connect to a URL? Run
uvx loopskill-mcp
instead — a small helper program (published on PyPI, the Python package index) that forwards
every call to the same hosted server. See
MCP setup for the config block.
Claude Code · Plugin marketplace (no key)
Add the LoopSkill marketplace to Claude Code
LoopSkill publishes a Claude Code plugin marketplace at wisechef-ai/loopskill-marketplace. Add it once, then install plugins by name. No signup and no key for the search plugin.
Inside a Claude Code session
/plugin marketplace add wisechef-ai/loopskill-marketplace
/plugin install loopskill-search@loopskill From a shell
claude plugin marketplace add wisechef-ai/loopskill-marketplace
claude plugin install loopskill-search@loopskill loopskill-search runs one
query across the curated catalog and the federated sources, so your agent can find an existing skill before
writing one. loopskill@loopskill
installs the LoopSkill skill file. Run /plugin
to browse the rest of the marketplace.
By hand · Skill file (no key)
Save the skill file yourself
This is what the one paste does for you. The LoopSkill skill is one Markdown file that tells your agent how to search and install skills. Save it in your agent's skills folder:
Claude Code
mkdir -p ~/.claude/skills/loopskill
curl -sL https://app.loopskill.io/skill -o ~/.claude/skills/loopskill/SKILL.md Hermes / Codex / OpenClaw
mkdir -p ~/.hermes/skills/loopskill # or ~/.codex/skills, ~/.openclaw/skills
curl -sL https://app.loopskill.io/skill -o ~/.hermes/skills/loopskill/SKILL.md Then in any session ask: "read the loopskill skill and follow it. List free skills." Your agent will hit the public catalog endpoints — no key needed.
For scripts · Plain HTTP
curl works too
Browse free skills (no auth)
curl -sS "https://app.loopskill.io/api/skills/search?tier=free&page_size=50" Install a free skill (no auth)
curl -sS "https://app.loopskill.io/api/skills/install?slug=super-memory" Install with a key (bundle-scoped or private skills)
curl -sS -H "x-api-key: $LOOPSKILL_API_KEY" \
"https://app.loopskill.io/api/skills/install?slug=super-memory" The catalog carries no paid-tier skills today, so no free skill actually needs this header — pass a key when installing from a private or bundle-scoped catalog instead.
Verify the install
curl https://app.loopskill.io/api/healthz
Should return {"status":"ok"}. If your agent connected via MCP, ask it: "call loopskill_search with query=test and show me the top 3 results."