jq, agents, and CI.
Pure JavaScript, Node 18+, no native binaries to download. Installs in under a second, starts in ~200 ms, ships as a single ~100 KB bundle.
Install
Requirements: Node.js 18+.- npm (recommended)
- One-liner (macOS / Linux)
- One-liner (Windows)
- No-install (npx)
olostep init is the recommended next step — it signs you in, installs the Olostep skills into your AI agents, and configures the MCP server, all in one command. The one-liner scripts wrap npm install -g olostep-cli with a Node 18+ check and a sudo fallback, so they work even if you’re unsure of your local setup.
Platforms: macOS (Apple Silicon and Intel), Linux (x64 and arm64), Windows (x64 and arm64).
Set up
One command does everything — sign in, install skills, and install the MCP server:--skills-only, --mcp-only, --no-browser, --relogin.
To just sign in (no skills/MCP):
olostep login):
Sign out
logout also warns you if OLOSTEP_API_KEY / OLOSTEP_API_TOKEN env vars or a .env file in your current directory still hold a key — those take priority over the credentials file, so deleting the file alone may not be enough. The output includes the exact unset commands for PowerShell and bash/zsh.
Quick start
--out <path> to save to a file.
What can it do?
Output
Every command prints its JSON result to stdout by default.
Progress and log lines go to stderr, so stdout stays clean for pipes.
search— you want a list of relevant URLs and snippets for a query. The CLI searches the web for you.answer— you want a synthesized answer, not raw page content. The CLI does the research for you.scrape— you already have the URL and want clean content out.crawl— you want every page on a site (or a filtered subset) without enumerating URLs by hand.batch-scrape— you have a list of URLs and want them processed in parallel.
Commands
Useolostep <command> --help for every option.
search: live web search
Returns deduplicated organic links (URL, title, description).
answer: researched answer
Synchronous — returns when the answer is ready.
map: discover URLs
scrape: one URL
Formats: html, markdown, text, json, raw_pdf, screenshot (comma-separated; default markdown).
scrape-get: fetch by ID
crawl: whole site
Starts a crawl, polls until finished, then retrieves page contents.
Retrieve formats: markdown, html, json.
Notable flags: --max-pages, --max-depth, --include-subdomain, --include-external, --include-url, --exclude-url, --search-query, --top-n, --webhook, --crawl-timeout, --formats, --pages-limit, --pages-search-query, --poll-seconds, --poll-timeout, --dry-run.
batch-scrape: CSV
CSV must have a header row with custom_id (or id) and url columns.
batch-update: batch metadata
Requires one of --metadata-json or --metadata-file (JSON object).
Auth commands
auth set-key is useful for CI and scripts — write the key directly without going through the browser flow.
Install the MCP server
The CLI writes the Olostep MCP server into your agent’s config — no JSON editing.
The hosted endpoint at
https://mcp.olostep.com/mcp uses Authorization: Bearer <key> — no local Node process required. The CLI merges only the olostep key into your existing config. Restart your agent after install.
Skills for AI agents
The CLI ships 13 Olostep skills —SKILL.md files installed into Claude Code, Cursor, and other agents so they know what Olostep can do and when to use it.
Health checks
Version & updates
Environment variables
Windows / PowerShell notes
PowerShell tokenizes, and * differently from bash — quote arguments:
$ interpolation).
See what’s installed
Global flags
--out, --timeout, and --api-key are available on every data command.
Security
Keep API keys out of source control; rotate if leaked.olostep logout removes the local credentials file and tells you if any env-var sources still hold a key.