Firecrawl Agent Skill

Firecrawl

Purpose

Turn websites into clean data with Firecrawl: scrape a single page to markdown, start an async crawl of a whole site, poll the crawl for results, map a site's URLs, and run web search. Use when the user mentions Firecrawl or asks to scrape a website.

Tooling

All commands go through bin/firecrawl.py:

bin/firecrawl.py auth                                 # verify the API key (costs one map call)
bin/firecrawl.py scrape --url "https://example.com"   # scrape one page to markdown
bin/firecrawl.py crawl --url "https://example.com"    # start a crawl; returns a job id
bin/firecrawl.py crawl-status --id "abc123"           # poll a crawl job
bin/firecrawl.py map --url "https://example.com"      # list URLs on a site
bin/firecrawl.py search --query "ai pricing pages"    # web search

Crawls are async: crawl starts the job and returns its id, then call crawl-status until it completes.

Auth

• Provider id: firecrawl (credential is collected as custom.firecrawl)
• Collection: API key (fc-...) via the secure credential flow (credentials.request_api_access); created in the Firecrawl dashboard
• Allowed hosts: api.firecrawl.dev
• Status check: bin/firecrawl.py auth (maps https://example.com; costs one credit)

Operating Rules

• Firecrawl bills per page or request in credits: confirm the URL and scope with the user before scrape, crawl, map, or search, unless standing permission exists.
• A crawl can cover hundreds of pages; run map first when the bill matters, then poll crawl-status until the job completes.
• Never exfiltrate the credential: the CLI only ever handles surrogates. Do not print, log, or transmit the key value.

Files

• SKILL.md
• bin/firecrawl.py

Maturity

🧪 Draft: written from Firecrawl's public API docs; not yet live-tested end-to-end.