Firecrawl

Firecrawl Universal Adapter

Source

• Upstream repository: https://github.com/firecrawl/firecrawl
• Local source checkout: %USERPROFILE%\.firecrawl-source\firecrawl
• Primary CLI package: firecrawl-cli
• Primary CLI command: firecrawl
• Optional MCP package: firecrawl-mcp
• Router delegator: skill-router firecrawl
• Universal stack source registry: %USERPROFILE%\.universal-ai-stack\config\source-integrations.json

Operating Model

Use Firecrawl as a shared hosted web-data CLI/API integration, not as copied skill bodies or a per-agent install. The universal install uses:

• one shallow upstream source checkout for README-grounded reference,
• one global npm CLI install of firecrawl-cli,
• this canonical skill for routing and workflow instructions,
• compact skill-router wrappers in each AI client,
• optional firecrawl-mcp only for clients that specifically need an MCP server.

No persistent MCP bridge is required for normal Firecrawl work. Use direct CLI or SDK/API calls first.

Safety And State Boundaries

Never commit or paste:

• FIRECRAWL_API_KEY,
• Firecrawl account, credit, billing, or workspace details,
• local auth/session files created by firecrawl login,
• project-local .firecrawl/ caches and generated outputs,
• generated scrape/crawl output from private or authenticated pages,
• browser profile data, cookies, screenshots, or downloaded private content.

Use the user's Firecrawl account only after they authenticate with firecrawl login, firecrawl login --browser, or a local FIRECRAWL_API_KEY. Respect robots.txt, site terms, auth boundaries, and rate limits. Ask before using authenticated pages, proxies, high-volume crawls, or batch scraping.

Prefer host-native web search for quick current lookups when available. Use Firecrawl when the task needs Firecrawl-specific search with scraped result content, hosted scraping, structured extraction, interact actions, URL mapping, site crawling, file parsing, or Firecrawl MCP/SDK integration.

Install And Health

Check local state:

skill-router firecrawl status

Install or update the CLI:

skill-router firecrawl install

Update the source checkout and reinstall the CLI:

skill-router firecrawl update

Authenticate when needed:

firecrawl login --browser

Check account/API readiness without exposing secrets:

firecrawl --status

Show optional MCP setup notes:

skill-router firecrawl mcp-help

CLI Workflows

Search with result content:

firecrawl search "model context protocol docs" --limit 5 --pretty

Scrape a page to clean Markdown:

firecrawl scrape https://example.com --only-main-content

Scrape with structured JSON:

firecrawl scrape https://example.com --format json --schema-file .\schema.json --pretty

Map a site:

firecrawl map https://example.com --json --pretty

Crawl a bounded site section:

firecrawl crawl https://example.com/docs --limit 25 --max-depth 2 --wait --pretty

Interact after a scrape:

firecrawl scrape https://example.com
firecrawl interact "Click the pricing link and summarize the plans"

Run an agent job:

firecrawl agent "Find the pricing plans for Notion" --pretty

Parse a local file through Firecrawl:

firecrawl parse .\document.pdf --format markdown

MCP Policy

Register firecrawl-mcp only when the target AI client needs live MCP tools. The portable config shape is:

{
  "mcpServers": {
    "firecrawl-mcp": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": {
        "FIRECRAWL_API_KEY": "from the user's local secret store"
      }
    }
  }
}

Do not add this MCP server globally by default. Router-first clients should use skill-router firecrawl ... and direct CLI calls.

When To Use This Skill

Use this skill for:

• Firecrawl CLI install, update, status, login, or diagnosis,
• Firecrawl hosted search, scrape, crawl, map, parse, interact, or agent jobs,
• Firecrawl Python/Node/cURL/API examples,
• Firecrawl MCP server configuration,
• Firecrawl source checkout or upstream README interpretation.

Reject this skill for:

• local Crawl4AI workflows; use crawl4ai,
• generic web search/current-news tasks that can use the host search tool,
• ordinary browser automation better handled by Browser, Playwright, or Lightpanda,
• unrelated "fire" or "crawl" wording that is not Firecrawl.