Search 41,000+ agent skills, MCP servers, Claude Code plugins, and agentic loops from your terminal. The CLI for getagentictools.com — real install counts and GitHub stars, refreshed weekly, never faked.
$ npx agentictools search pdf extraction
pdf-extraction [skills] ↓ 7.8k skills/claude-office-skills-skills-pdf-extraction
pdf extraction
PDFMux [mcp] ↓ 15k mcp/nameetp-pdfmux
Universal PDF extraction orchestrator that routes pages to the best backend…npx agentictools search <query> # search everything
npx agentictools search browser -c mcp # one category: skills | mcp | plugins | loops
npx agentictools info skills/mattpocock-skills-tdd
npx agentictools install <id> # shows the install command — never executes it
npx agentictools open <id> # its page, in your browser
npx agentictools statsNo install needed — npx runs it. Zero dependencies, Node ≥ 20.
| Command | What it does |
|---|---|
search <query> |
Ranked search across the catalog. Flags: -c/--category, -n/--limit, --json, --refresh |
info <id> |
Full detail: metrics, author, license, last upstream push, install command, page URL |
install <id> |
Prints the install command and where it runs — your shell, or inside a Claude Code session for slash commands. Never executes. |
open <id> |
Opens the tool's catalog page (prints the URL when piped) |
stats |
Catalog counts + which deploy the data came from |
IDs are as search shows them: bare (mattpocock-skills-tdd), or qualified (skills/mattpocock-skills-tdd) when a bare id exists in more than one category.
Half the catalog installs via Claude Code slash commands (/plugin marketplace add …) that aren't shell commands at all, and the rest are third-party commands you should see before running. install shows you exactly what to run and where — the honest version of one-click.
Every command takes --json and prints results to stdout only (status goes to stderr), so output pipes cleanly:
npx agentictools search "sql database" -c mcp --json | jq -r '.[0].url'
npx agentictools install some-tool --json | jq '{installCommand, kind}'Add it to your agent's toolbox: an agent that can run npx agentictools search --json <query> can discover skills, MCP servers, plugins, and loops on demand. The --json shapes are documented in AGENTS.md.
Your agent speaks MCP? The same catalog is a free remote MCP server — one command to connect:
claude mcp add --transport http getagentictools https://getagentictools.com/api/mcpThe catalog comes from getagentictools.com — harvested from GitHub and public registries, refreshed weekly. Metrics are real installs and stars, never entered by hand. Browse it on the web: skills · MCP servers · plugins · loops
No telemetry. No accounts. The only network call is one JSON fetch of the catalog index, cached locally (~/.cache/agentictools) for 24 hours with ETag revalidation. --refresh revalidates now; offline runs use the cache.
Every catalog page shows its own CLI line and badge — copy them into your README:
`npx agentictools info <category>/<your-tool-id>`
[Listed on getagentictools](https://getagentictools.com?ref=badge)MIT © Daniel Moka