Firecrawl ships its web-scraping brain as terminal commands and agent skills

What it does

The Firecrawl CLI wraps the Firecrawl web data API in terminal commands: scrape a URL to clean markdown, search the web (including news, images, research-affiliated sites, developer sources like GitHub issues and PRs, and US government sources), crawl sites, map them, and run browser interactions. It also installs “skills” into AI coding agents — Claude Code, Cursor, Codex, Windsurf, and several others — teaching those agents to use the CLI for live web work, plus optional workflow skills for deliverables like research briefs, SEO audits, and lead lists.

The interesting bit

The skills system is the real story. Rather than just exposing an MCP server, the CLI can install instruction-based skills into every coding agent it detects on your machine, and can even make Firecrawl the default web provider — disabling agents’ native web fetch/search so they route web work through Firecrawl instead. It’s a distribution play: the terminal is the product, the agents are the customers.

Key highlights

• Scraping supports many output formats — markdown, cleaned HTML, links, images, screenshots, AI summaries, structured JSON extraction via schema, change tracking, and brand identity extraction.
• Multiple URLs are scraped concurrently, with results auto-saved to .firecrawl/.
• Works against self-hosted Firecrawl instances via --api-url; local instances need no API key.
• --redact-pii and --lockdown flags exist for privacy-conscious scraping.
• Skills can be scoped per-harness (--agent claude-code, --agent codex, etc.) instead of installed everywhere.

Caveats

• This is a client for a commercial service — the value lives server-side, and the CLI is (well-made) glue over Firecrawl’s API.
• The README is heavy on setup and command listings; details on pricing, rate limits, and how the agent skills actually perform aren’t covered here.
• The setup defaults feature modifies your agents’ existing web-fetch configuration — reversible via --undo, but worth knowing before running it with -y.

Verdict

Worth a look if you use an AI coding agent and want it to do real web research instead of hallucinating citations. If you’re avoiding vendor-locked scraping services, this repo won’t change your mind — the open-source value is in the skills and CLI ergonomics, not the engine.