Skills
v1.2.0Search a billion+ documents — papers, books, code, legal cases, forums, Wikipedia, and more.
Like a lobster shell, security has layers — review code before you run it.
License
Runtime requirements
SKILL.md
JAIL Search
Search a billion+ documents — papers, books, code, legal cases, forums, Wikipedia, and more. Returns titles, authors, URLs, and short descriptions — for discovering sources and links, not retrieving full content. After finding relevant results, use their URLs with fetch or browsing tools to read the actual documents.
When to Use
- User asks to research a topic, find papers, books, or articles
- User wants to look up facts, people, places, or concepts
- User asks about community discussions or forum threads
- User wants to find legal cases or legislation
- User needs to find music, movies, packages, or other metadata
- User asks "search for...", "find...", or "look up..."
Via MCP (preferred if jail MCP server is configured)
Use mcp_jail_search(query, type) and mcp_jail_detail(doc_id).
Setup (if not already configured)
Connect to the hosted MCP server:
claude mcp add --transport http jail https://api.jail.li/mcp
With API key for higher rate limits:
claude mcp add --transport http jail "https://api.jail.li/mcp?jailApiKey=sk_live_your_key_from_jail_li"
For other clients, add to MCP config: "url": "https://api.jail.li/mcp?jailApiKey=sk_live_..."
Get a key at jail.li.
Via curl (fallback)
Search
curl -Gs "https://api.jail.li/v1/search" --data-urlencode "q=QUERY" -d "type=TYPE&limit=10"
With API key: add -H "Authorization: Bearer $JAIL_API_KEY" or use ?jailApiKey=sk_live_...
Replace QUERY and TYPE (required, see below).
Paginate: add &cursor=CURSOR using next_cursor from previous response.
Detail
curl -s "https://api.jail.li/v1/detail/DOC_ID"
Types
Start with: academic, wiki, books, legal, forums. The rest just exist if you need.
| Type | Content |
|---|---|
academic | OpenAlex, arXiv, Semantic Scholar, DBLP |
wiki | Wikipedia (18 languages) |
books | Books, digital libraries, and classical literature |
legal | Harvard Case Law, CourtListener, EUR-Lex, UK Legislation |
forums | Hacker News, StackExchange, Lobsters, LessWrong, and 60+ more |
economics | World Bank, IMF, FRED, ECB, BLS, Tax Foundation |
packages | npm, PyPI, Crates.io, Libraries.io |
knowledge | Wikidata, structured knowledge, and facts |
news | News articles and journalism |
music | Discogs, MusicBrainz |
video | IMDb, YouTube |
health | Clinical trials and food safety data |
geo | World place names and geographic data |
fandom | Fan wiki articles and community knowledge bases |
tech | Dev.to, product community forums |
audio | Podcasts and audio content |
social | Mastodon, Lemmy, fediverse |
crypto | DeFi protocols, token data, and on-chain analytics |
predictions | Prediction markets and forecasting |
Strategy
- Pick the right type first — this determines which indices are searched
- Use 2-4 keywords (English preferred unless searching non-English content)
- Try different keywords and synonyms if first attempt returns few results
- Search the same topic across multiple types for cross-referencing
- Use
next_cursorto paginate for more results - Use detail endpoint for full metadata on promising results
Response fields
Each result: title, author, year, type, description (200 char), id, url, score.
Get an API key at https://jail.li for higher limits.
Files
1 totalComments
Loading comments…
