---
title: Firecrawl CLI
category: product
entity_type: skill
price: $15
canonical: https://forgehouse.ai/skills/firecrawl-cli/
lang: en
hreflang_alt: https://forgehouse.ai/tr/skiller/firecrawl-cli/
last_updated: 2026-06-20
---

# Firecrawl CLI

> Official Firecrawl CLI skill for web scraping, search, crawling, and browser automation.

A disciplined command-line workflow for web scraping, search, crawling, and browser automation that returns clean, LLM-ready markdown instead of messy HTML. It enforces a five-step escalation pattern: search, scrape, map, crawl, browser, so you never burn API credits or hammer a server when a lighter tool would do. Built around credit tracking, rate limiting, and robots.txt compliance so your data collection stays fast, polite, and legal.

## Use cases
- Researching a topic and scraping the top results in one pass
- Extracting clean markdown from articles and documentation pages
- Mapping a large site to locate a specific subpage before scraping
- Bulk-downloading an entire docs section into local files
- Pulling structured data (pricing tiers, product lists) from complex sites
- Reaching content behind pagination, modals, or login via cloud browser sessions

## Benefits
- Stop wasting API credits: escalation logic and credit checks prevent redundant fetches
- Get content your model can actually read, not raw HTML you have to clean
- Avoid IP bans and legal exposure through built-in crawl politeness and robots.txt handling
- Run independent scrapes in parallel up to your concurrency limit for faster collection

## What’s included
- Five-command toolkit: search, scrape, map, crawl, and cloud browser automation
- Credit-usage and status checks to verify auth and remaining budget before any job
- Crawl controls for delay, max-concurrency, depth limits, and path include/exclude filters
- AI agent extraction with JSON schema support for structured output
- Bulk download command combining map plus scrape to save whole sites locally
- Output organization conventions with jq/grep patterns for working with large result files

## Who it’s for
Developers, researchers, and AI engineers who need to feed clean, structured web content into LLM pipelines without managing scraping infrastructure.

## How it runs
Scraping escalates from cheapest call to full browser session, never the other way. Auth and credits get checked first, crawls respect politeness limits, results land in files read incrementally, and counts are verified before the job is called done.
1. Gate check before anything fires: runs firecrawl --status to verify auth and remaining credits. An invalid API key returns silent empty JSON instead of an error, so this check always comes first.
2. Escalates through five tiers and never skips ahead: no URL yet means search (with --scrape to pull full page content in the same call), a known URL means scrape with --only-main-content, a large site means map --search to locate the exact subpage first.
3. Bulk extraction runs crawl with --include-paths, --limit, --delay and --max-concurrency set deliberately for crawl politeness: small sites get 500ms delay at concurrency 2, robots.txt is checked before any client-site crawl.
4. Browser is the last resort, used only when scrape fails behind interaction: open the page, snapshot to get @ref element IDs, click or fill by ref, then scrape the rendered state. Never used for web search.
5. Every result is written to .firecrawl/ with a naming convention and read incrementally via grep and head, never loaded whole. search --scrape output is reused instead of re-scraping the same URLs, which burns credits twice.
6. Output is verified before continuing: result counts checked with jq and wc -l. If the file holds 20 percent fewer pages than expected, that signals a rate limit or auth failure, not a finished job.

## FAQ
### Do I need a Firecrawl account and key to use this?
Yes, it is a command-line workflow over the Firecrawl service, so it expects a key. The discipline is about spending those credits well, escalating only when a lighter call cannot answer.

### Why bother with five steps, why not just crawl the whole site every time?
Crawl is the heaviest call, and a search or single scrape usually gets the answer for a fraction of the cost. The escalation order exists so you stop hammering servers and burning credits on work a lighter step already does.

### Will it scrape pages behind a login or strong bot protection?
Browser automation handles JavaScript-rendered pages, but it is not built to defeat hard authentication or aggressive anti-bot walls. Where a site blocks automated access, this respects that rather than forcing through.

## Price
$15, one-time, no subscription. VAT included.

Related guide: [AI for data analytics](https://forgehouse.ai/guides/ai-data-analytics/)