Overview
Firecrawl is a hosted web-data infrastructure service — a “context API to
search, scrape, and interact with the web at scale” that turns websites into
clean, LLM-ready data (markdown, JSON, or screenshots). It is the company
behind the firecrawl-deep-research skill evaluated in
is-firecrawl-useful-to-this-wiki-purpose and the publisher of the
firecrawl-workflows skill pack for AI coding agents. Both captured sources
are vendor-authored marketing material, so capability claims below carry no
independent verification.
Key Facts
- Three core capabilities: Search (query → ranked results with full-page markdown), Scrape (URL → structured data, with JS rendering for client-heavy pages), and Interact (operate a page: click, type, scroll — for login-gated or paginated content). Also offers Crawl (site-wide), Map, Parse (PDF/DOCX), and Monitor.
- Vendor market claims: open-source core with 160.9K GitHub stars (a top-100 repo), 1.25M developers, 150K+ companies (listed customers include Apple, Canva, Lovable, Zapier, Replit), 5B+ requests served, 400K+ MCP server installs, SOC 2 Type 2. None independently verified in this vault.
- Vendor performance claims: P95 latency 3.4s, ~93% fewer input tokens via clean markdown (no nav/footer/ads), ~96% web coverage including JS-heavy pages.
- Pricing: free 1,000 credits/month; Scrape/Crawl/Map/Monitor = 1 credit/page, Search = 2 credits/10 results, Interact = 2 credits/browser-minute.
- Distribution: official MCP server, CLI (
npx -y firecrawl-cli@latest init --all --browser), and agent skills for Claude Code/Codex. - The open-source core lacks Interact; the hosted product runs proprietary “Fire-engine” infrastructure (proxies, rendering).
Firecrawl Workflows
firecrawl/firecrawl-workflows (ISC license, ~110 stars) ships 16
outcome-focused skills for AI coding agents — including deep-research,
seo-audit, qa, lead-research, competitive-intel, knowledge-base,
knowledge-ingest, and website-design-clone — split into Core (live web
work via CLI), Build (integrate into app code), and Workflows (repeatable
business/creative deliverables). Its authoring principles (harness-agnostic,
infer-from-context with at most 1–3 blocking questions, target a real
deliverable, structure for recurring automation) borrow the layout of
anthropics/skills and are reusable for this vault’s own skill design
regardless of the service itself.
Relationships
- “implements” model-context-protocol — official MCP server (400K+ installs per vendor claim)
- “references” is-firecrawl-useful-to-this-wiki-purpose — subject of the vault’s filed tool-fit verdict (low-to-medium utility)
- “differs from” external-links-connections-deep-dive — hosted SaaS capture vs the vault’s local
external_links.pypipeline - “compatible-with” claude-skills — ships agent skills for Claude Code/Codex
Implications
Firecrawl is a capable hosted capture backend, but it duplicates the vault’s
existing scripts/external_links.py capture pipeline and conflicts with the
local-first governance in vault-roadmap. It stays an optional “power
capture” backend for JS-rendered or anti-bot pages only, with output still
funneled through the capture pipeline before it becomes evidence (Guardrail
#16). The more durable takeaway is the workflows pack’s skill-design
principles, which are usable here independent of the service.
Sources
^[raw/external/www-firecrawl-dev-a13932d6.md]