Command-line interface for Firecrawl, the context API for AI agents. Search, scrape, parse, crawl, map, and interact with the web, search research papers, developer sources, and government sources, and run agent jobs directly from your terminal.

npm install -g firecrawl-cli

Or set up everything in one command (install CLI globally, authenticate, and install shared skills for your detected coding editors):

npx -y firecrawl-cli@latest init -y --browser

• -y runs setup non-interactively
• --browser opens the browser for Firecrawl authentication automatically
• skills install globally in the shared skills directory and are made available to detected AI coding agents by default

If you are using an AI coding agent like Claude Code, you can also install skill groups manually — one command per family:

firecrawl setup core       # scrape/search/crawl/interact primitives + index skills ("skills" is an alias)
firecrawl setup build      # app-integration skills for the Firecrawl API
firecrawl setup workflows  # end-to-end recipes (lead gen, deep research, ...)

Or install a single skill by name — the firecrawl- prefix is optional:

firecrawl setup developer-index
firecrawl setup seo-audit

If no API key is found afterwards, an interactive terminal offers a browser login (pass --browser to log in without the prompt); non-interactive runs never block — they print a hint and the skills walk agents through setup on first use.

Default installs use the shared ~/.agents/skills directory and the coding editors detected by Firecrawl. The npx installer always includes Codex as the shared-directory target, even when Codex is not detected, preserving shared files and agent links. If no editors are detected, only the shared directory is populated. The npx path detects supported global targets, including Kiro and custom configuration locations such as CLAUDE_CONFIG_DIR and XDG_CONFIG_HOME. Project-only targets are excluded. Use --agent <agent> to target a specific editor.

By default skills route to every detected harness. To install and route to only one harness in a single command:

# One-shot from the install script (non-interactive, no TTY needed)
curl -fsSL https://firecrawl.dev/install.sh | bash -s -- --agent openclaw

# Or, once the CLI is installed
firecrawl init --agent openclaw          # only OpenClaw
firecrawl setup skills --agent openclaw
firecrawl setup workflows --agent openclaw

Any args after bash -s -- are forwarded to firecrawl init, so you can add things like --skip-auth for a skills-only, no-login setup. When you run firecrawl init interactively without --agent, it shows a checkbox of your detected harnesses (all selected by default) so you can pick a subset.

The init command installs the CLI skills by default and offers the workflow skills as optional extras, into AI coding agents (Cursor, Claude Code, Windsurf, etc.):

• CLI skills — teach agents how to use the Firecrawl CLI for live web work (search, scrape, interact, map, crawl, agent) and Alexandria data providers. Installed by default.
• Workflow skills — teach agents how to produce Firecrawl-powered deliverables such as research briefs, SEO audits, QA reports, lead lists, knowledge bases, and design-system extraction. Interactive multi-select during init.

All skill families live in the firecrawl/skills catalog — including the build skills for integrating Firecrawl into application code:

npx skills add firecrawl/skills

Contributing skills? CLI skills (including the research/developer index skills) → PR this repo (skills/). Build/SDK skills → PR the firecrawl monorepo (skills/). Workflow skills → PR firecrawl/firecrawl-workflows. The catalog (firecrawl/skills) is read-only — never PR it directly.

The Firecrawl Claude Code and Cursor plugins are released separately and do not include these CLI skills. Changes here still reach the skill catalog.

To reinstall skills manually:

firecrawl setup core
firecrawl setup build
firecrawl setup workflows

Global npm, pnpm, bun, and yarn installations, plus standalone binaries on macOS and Linux, check for updates in the background. Commands do not wait for an update, and the updater never writes to stdout. The next invocation uses the updated release, with a one-time notice on an interactive terminal.

For global package installs, the installed package is a stable launcher. The updater uses npm to install a new release into a separate directory under ~/.firecrawl/versions/, verifies it, then atomically switches the current version pointer. It never rewrites the global installation or removes files used by running commands. Managed versions are shared within a Node major version, so a patch upgrade reuses the saved release. The launcher checks the release's Node requirements before selecting it. npm must be available on PATH for automatic package updates; otherwise use your package manager's manual update command. Install scripts are disabled. Dependencies resolve to versions published no later than the release itself, so the waiting period covers them too.

Each global installation has its own version selection. Maintenance retains the current and previous release for each installation, plus any older release still used by a running CLI process. It removes other inactive versions and staging directories older than 24 hours whose installer has exited. Files owned by another machine on shared storage are retained conservatively.

Reinstalling with npm resets the launcher's selection to that package, including an explicit downgrade. Disable automatic updates before reinstalling if you want to stay on the selected version. Disabling updates alone keeps the currently selected release.

Standalone binaries download their release archive, verify its SHA-256 checksum and reported version, then atomically replace the executable. The previous binary is retained beside it with a .previous suffix.

A newly observed release waits at least 24 hours before unattended installation. The worker rechecks that it is still the latest release after this delay; a changed candidate starts a new waiting period. Version checks normally refresh about once a day. Automatic updates never cross a major version or install a prerelease. Prerelease installations remain pinned.

To update immediately, bypassing the waiting period, including to a new major version:

firecrawl update

To restore the previous version without the network:

firecrawl update --rollback

For global package installs, the launcher handles rollback without loading the active release, so it works even when that release cannot start. Rollback disables automatic updates so the same release is not immediately installed again. The first managed-package rollback returns to the original package-manager installation; later updates can roll back to the previous managed version. Moving the registry's latest tag back does not automatically downgrade an existing installation.

Control automatic updates with:

firecrawl update --disable-auto
firecrawl update --enable-auto
export FIRECRAWL_NO_AUTO_UPDATE=1

Automatic updates are also off when CI is set, unless it is 0 or false. FIRECRAWL_NO_UPDATE_CHECK=1 disables both automatic updates and notices. Explicit updates still work with these settings. Failures leave the active installation unchanged and back off before retrying. Details are recorded in ~/.firecrawl/auto-update.log.

Homebrew and other system packages, project-local installs, package runners such as npx, and standalone Windows binaries retain their selected release and show the appropriate manual update instruction. Enabling automatic updates on these installations reports that they are externally managed. Disabling updates saves the shared user preference from any installation. Non-writable or differently owned standalone binaries are not replaced.

Package updates use npm's registry configuration. Binary version checks honor npm_config_registry; binary downloads use GitHub releases, or the mirror set with FIRECRAWL_RELEASES_URL.

Interactive CLI sessions (except init, setup, and the -y/--yes init shorthand) show a boxed notice when installed Firecrawl skills differ from the current catalog. The check includes supporting reference files and runs independently of CLI releases. Installs record which catalog skills existed, so the notice announces skills published later, once, in families whose umbrella skill is installed. Skills skipped during setup are not offered again.

Each affected family gets one command: firecrawl setup core, firecrawl setup build, or firecrawl setup workflows. These commands install the entire current family, including previously skipped skills and new skills added since your CLI was released. When installing through npx, failed catalog API discovery falls back to this CLI release's known names with a warning. Native installs discover skills directly from the Git repository and require that repository to be available. An explicit refresh replaces installed files, including local edits; the notice itself never changes them.

Checks cover global installations recorded in ~/.agents/.skill-lock.json with files in ~/.agents/skills. They cache catalog results for 20 hours, repeat the notice at most once every 12 hours, and retry failed checks after an hour. Non-interactive runs stay quiet. Set FIRECRAWL_NO_UPDATE_CHECK=1 to disable both CLI and skill update notices.

To install the Firecrawl MCP server into your editors (Cursor, Claude Code, VS Code, etc.):

firecrawl setup mcp

With a key saved by firecrawl login, export FIRECRAWL_API_KEY and pass --agent <agent> (or --agent all), or add --keyless: setup never writes a stored key into client configs.

Use --agent <agent> for one client, or --agent all --global to configure Claude Code, VS Code, Codex, OpenCode, Cursor, Gemini CLI, Windsurf, Antigravity, Amp, Grok Build, Zed, Hermes Agent, and OpenClaw in one run. The --agent values for the newer clients are amp (or ampcode, amp-code), grok (or grok-build, grok-cli), and zed. Amp, Grok Build, and Zed are written to their user settings, so they do not take --project: Amp to ~/.config/amp/settings.json (or an existing settings.jsonc when there is no settings.json), Grok Build to ~/.grok/config.toml, and Zed to ~/.config/zed/settings.json ($XDG_CONFIG_HOME/zed/settings.json on Linux when it is set, %APPDATA%\Zed\settings.json on Windows). Clients that cannot be configured on this machine (for example OpenClaw when the openclaw command is not installed) are skipped and listed in the summary; the command only fails when a client it tried to configure fails. When FIRECRAWL_API_KEY is set, Antigravity and Zed are skipped because they have no verified way to reference the key from the environment; add --keyless to include them.

To make Firecrawl the default web provider for supported AI agents:

firecrawl setup defaults    # or: firecrawl make default

This disables native web fetch/search where supported so agents route web work through Firecrawl. When run interactively, it asks harness by harness (Claude Code, Codex) so you can choose exactly which to change. Use -y to skip the picker and apply to all, or --agent to target one:

firecrawl setup defaults --agent codex      # only Codex
firecrawl setup defaults -y                  # all harnesses, no prompts

To undo those config changes (also interactive, harness by harness):

firecrawl setup defaults --undo              # pick which to restore
firecrawl setup defaults --undo --agent claude

Just run a command - the CLI will prompt you to authenticate if needed:

firecrawl https://example.com

On first run, you'll be prompted to authenticate:

  🔥 firecrawl cli
  Search, scrape, and interact with the web

Welcome! To get started, authenticate with your Firecrawl account.

  1. Login with browser (recommended)
  2. Enter API key manually

Tip: You can also set FIRECRAWL_API_KEY environment variable

Enter choice [1/2]:

# Interactive (prompts automatically when needed)
firecrawl

# Browser login
firecrawl login

# Direct API key
firecrawl login --api-key fc-your-api-key

# Environment variable
export FIRECRAWL_API_KEY=fc-your-api-key

# Per-command API key
firecrawl scrape https://example.com --api-key fc-your-api-key

For self-hosted Firecrawl instances or local development, use the --api-url option:

# Use a local Firecrawl instance (no API key required)
firecrawl --api-url http://localhost:3002 scrape https://example.com

# Or set via environment variable
export FIRECRAWL_API_URL=http://localhost:3002
firecrawl scrape https://example.com

# Self-hosted with API key
firecrawl --api-url https://firecrawl.mycompany.com --api-key fc-xxx scrape https://example.com

When using a custom API URL (anything other than https://api.firecrawl.dev), authentication is automatically skipped, allowing you to use local instances without an API key.

Extract content from any webpage. Pass multiple URLs to scrape them concurrently -- each result is saved to .firecrawl/ automatically.

# Basic usage (outputs markdown)
firecrawl https://example.com
firecrawl scrape https://example.com

# Get raw HTML
firecrawl https://example.com --html
firecrawl https://example.com -H

# Multiple formats (outputs JSON)
firecrawl https://example.com --format markdown,links,images

# Save to file
firecrawl https://example.com -o output.md
firecrawl https://example.com --format json -o data.json --pretty

# Multiple URLs (scraped concurrently, each saved to .firecrawl/)
firecrawl scrape https://firecrawl.dev https://firecrawl.dev/blog https://docs.firecrawl.dev

# Extract only main content as markdown
firecrawl https://blog.example.com --only-main-content

# Wait for JS to render, then scrape
firecrawl https://spa-app.com --wait-for 3000

# Get all links from a page
firecrawl https://example.com --format links

# Screenshot + markdown
firecrawl https://example.com --format markdown --screenshot

# Extract specific elements only
firecrawl https://example.com --include-tags article,main

# Exclude navigation and ads
firecrawl https://example.com --exclude-tags nav,aside,.ad

Search the web with query-relevant highlights and optionally scrape content from search results.

# Basic search
firecrawl search "firecrawl"

# Limit results
firecrawl search "AI news" --limit 10

# Search news sources
firecrawl search "tech startups" --sources news

# Search images
firecrawl search "landscape photography" --sources images

# Multiple sources
firecrawl search "machine learning" --sources web,news,images

# Filter by category (research-affiliated websites, PDFs, developer index, gov index)
firecrawl search "transformer architecture" --categories research
firecrawl search "machine learning" --categories pdf,research

# Note: --categories research narrows *web* results to research-affiliated
# websites. To search papers themselves, use `firecrawl research search-papers`.

# Developer search: public repositories, GitHub issues, merged PRs, READMEs, and docs
firecrawl search "axum middleware ordering" --categories developer

# Government search: US government sources (cannot be combined with other categories)
firecrawl search "California data breach notification statute" --categories gov

# Time-based search
firecrawl search "AI announcements" --tbs qdr:d   # Past day
firecrawl search "tech news" --tbs qdr:w          # Past week

# Location-based search
firecrawl search "restaurants" --location "San Francisco,California,United States"
firecrawl search "local news" --country DE

# Search and scrape results
firecrawl search "firecrawl tutorials" --scrape
firecrawl search "API documentation" --scrape --scrape-formats markdown,links

# Output as pretty JSON
firecrawl search "AI data tools"

# Research a topic with recent results
firecrawl search "React Server Components" --tbs qdr:m --limit 10

# Search public developer sources
firecrawl search "web data library" --categories developer --limit 20

# Search and get full content
firecrawl search "firecrawl documentation" --scrape --scrape-formats markdown --json -o results.json

# Find research papers -- use the paper index, not the website filter
firecrawl research search-papers "large language models" --json

# Narrow web results to research-affiliated websites (not the paper index)
firecrawl search "large language models" --categories research --json

# Answer a programming question from public repositories, GitHub issues, merged PRs, READMEs, and docs
firecrawl search "tokio select cancellation safety" --categories developer --json

# Search with location targeting
firecrawl search "best coffee shops" --location "Berlin,Germany" --country DE

# Get news from the past week
firecrawl search "AI startups funding" --sources news --tbs qdr:w --limit 15

Search an index built for coding agents: public repositories, GitHub issues, merged pull requests, repository READMEs, and curated documentation sites. Use it for a programming question: code behaviour, a library or framework, an API contract, an error message, or a known bug.

The CLI intentionally keeps this agent-facing surface lean: it accepts only the query and result count. Express repository, source, result-kind, language, topic, license, and other scoping intent in the query text; semantic retrieval handles the scoping. For advanced filters, use the Developer Index REST API.

firecrawl developer "axum middleware ordering"

# Investigate a known bug
firecrawl developer "tokio spawn_blocking panics thread limit" --limit 10

# Put repository and evidence-kind intent directly in the semantic query
firecrawl developer "tokio select cancellation safety in tokio-rs/tokio issues and merged pull requests"

# Keep the full response, including passages, citations, and licenses
firecrawl developer "tokio select cancellation safety" --json -o results.json

Search the Government Index: primary law and regulatory material from US federal, state, and local government sources, including statutes, regulations, codes, court opinions, and other government publications.

For the request and response schema, see the Government Index REST API.

firecrawl gov "food labeling requirements for allergens"

Filters combine with AND. Filters keep only pages the index has labeled, so a filtered search can leave out relevant pages, and no results does not mean none exist. Court labels are most complete for federal courts and case-law collections.

# Find state statutes on a topic
firecrawl gov "California data breach notification statute" --limit 10

# Narrow to one court
firecrawl gov "qualified immunity excessive force" --court ca9

# Keep only pages that cite a statute
firecrawl gov "civil rights claims against police officers" --cites "42 U.S.C. § 1983"

# Keep the raw response
firecrawl gov "FDA food labeling regulations" --json -o results.json

Search Firecrawl's research paper index: roughly 43M abstracts, around 90% biomedical (PubMed, bioRxiv, medRxiv) plus arXiv. Use this for biomedical, clinical, and scientific literature rather than scraping PubMed, bioRxiv, or Google Scholar by hand.

This is not the same thing as firecrawl search --categories research, which only narrows ordinary web results to research-affiliated websites.

# Find papers by topic (start here)
firecrawl research search-papers "CRISPR base editing off-target effects" --limit 20

# Full metadata for one paper, by any supported id form
firecrawl research inspect-paper pmid:40953549

# Expand along the citation graph from your strongest hits
firecrawl research related-papers pmcid:PMC12530322 --intent "in vivo delivery"

# Read full-text passages to verify a specific claim
firecrawl research read-paper doi:10.1016/j.neunet.2025.108095 --question "What was the sample size?"

Paper ids accept pmid:, pmcid:, doi:, and arxiv: forms, plus canonical paperId values returned by search.

• Run several distinct framings of the same question rather than one query -- recall improves markedly.
• Use search-papers to find anchors, related-papers to expand, then read-paper to verify a candidate before including it.

Send concise feedback for a completed v2 search, scrape, parse, or map job. For search-result quality, search-feedback is still the most guided command; feedback is the generic endpoint/job surface.

firecrawl feedback scrape 0193f6c5-1234-7890-abcd-1234567890ab \
  --rating partial \
  --issues missing_markdown \
  --tags docs \
  --note "The pricing table was missing from the markdown output." \
  --url https://example.com/pricing \
  --page-numbers 1

Keep notes and metadata small. Do not send raw scrape or parse outputs as feedback.

Set FIRECRAWL_NO_ENDPOINT_FEEDBACK=1 to make firecrawl feedback skip endpoint feedback calls silently.

Quickly discover all URLs on a website without scraping content.

# List all URLs (one per line)
firecrawl map https://example.com

# Output as JSON
firecrawl map https://example.com --json

# Search for specific URLs
firecrawl map https://example.com --search "blog"

# Limit results
firecrawl map https://example.com --limit 500

# Find all product pages
firecrawl map https://shop.example.com --search "product"

# Get sitemap URLs only
firecrawl map https://example.com --sitemap only

# Save URL list to file
firecrawl map https://example.com -o urls.txt

# Include subdomains
firecrawl map https://example.com --include-subdomains --limit 1000

Crawl multiple pages from a website.

# Start a crawl (returns job ID)
firecrawl crawl https://example.com

# Wait for crawl to complete
firecrawl crawl https://example.com --wait

# With progress indicator
firecrawl crawl https://example.com --wait --progress

# Check crawl status
firecrawl crawl <job-id>

# Limit pages
firecrawl crawl https://example.com --limit 100 --max-depth 3

# Crawl blog section only
firecrawl crawl https://example.com --include-paths /blog,/posts

# Exclude admin pages
firecrawl crawl https://example.com --exclude-paths /admin,/login

# Crawl with rate limiting
firecrawl crawl https://example.com --delay 1000 --max-concurrency 2

# Deep crawl with high limit
firecrawl crawl https://example.com --limit 1000 --max-depth 10 --wait --progress

# Save results
firecrawl crawl https://example.com --wait -o crawl-results.json --pretty

Scrape many known URLs as one async job. Unlike multi-URL scrape, every requested format is kept and results come back as JSON.

# Start a batch scrape (returns job ID)
firecrawl batch-scrape https://example.com https://example.org

# Read URLs from a file (one per line; blank lines and # comments ignored)
firecrawl batch-scrape --urls-file urls.txt

# Wait for results
firecrawl batch-scrape --urls-file urls.txt --format markdown,links --wait --progress

# Check status (returns documents scraped so far)
firecrawl batch-scrape <job-id>

# Cancel a job
firecrawl batch-scrape <job-id> --cancel

# Extract the same fields from every page
firecrawl batch-scrape --urls-file products.txt --wait \
  --schema '{"type":"object","properties":{"name":{"type":"string"},"price":{"type":"string"}}}' \
  -o products.json --pretty

# Notify a webhook instead of waiting
firecrawl batch-scrape --urls-file urls.txt --webhook https://example.com/hooks/firecrawl

# Show credit usage
firecrawl credit-usage

# Output as JSON
firecrawl credit-usage --json --pretty

Run an AI agent that autonomously browses and extracts structured data from the web based on natural language prompts.

Note: Agent tasks typically take 2 to 5 minutes to complete, and sometimes longer for complex extractions. Use sparingly and consider --max-credits to limit costs.

# Basic usage (returns job ID immediately)
firecrawl agent "Find the pricing plans for Firecrawl"

# Wait for completion
firecrawl agent "Extract all product names and prices from this store" --wait

# Focus on specific URLs
firecrawl agent "Get the main features listed" --urls https://example.com/features

# Use structured output with JSON schema
firecrawl agent "Extract company info" --schema '{"type":"object","properties":{"name":{"type":"string"},"employees":{"type":"number"}}}'

# Load schema from file
firecrawl agent "Extract product data" --schema-file ./product-schema.json --wait

# Check status of an existing job
firecrawl agent <job-id>
firecrawl agent <job-id> --wait

# Research task with timeout
firecrawl agent "Find the top 5 competitors of Notion and their pricing" --wait --timeout 300

# Extract data with cost limit
firecrawl agent "Get all blog post titles and dates" --urls https://blog.example.com --max-credits 100 --wait

# Spend more reasoning on a complex extraction
firecrawl agent "Extract detailed technical specifications" --effort high --wait --json --pretty

# Save structured results to file
firecrawl agent "Extract contact information" --schema-file ./contact-schema.json --wait --json -o contacts.json --pretty

# Check job status without waiting
firecrawl agent abc123-def456-... --json

# Poll a running job until completion
firecrawl agent abc123-def456-... --wait --poll-interval 10

A run uses Alexandria providers only when it starts with an Alexandria flag. A follow-up turn (--thread) keeps the previous turn's setting for every Alexandria flag it omits.

# Let the agent use two providers
firecrawl agent "Find the head of sales at example.com" --toolkits apollo,crunchbase --wait

# Ask before any paid provider call, then answer the approval on the next turn
firecrawl agent "Find the head of sales at example.com" --mode chat --require-approval --wait
firecrawl agent "Go ahead" --thread <thread-id> --mode chat --approve <approval-id> --wait

A run that stops for approval prints its approval ID and the exact --approve / --decline follow-up commands. With --json, read pendingApproval and exchange from the output.

Scrape a page, then interact with it in a live browser session using natural language or code. No manual session management required.

# 1. Scrape a page first
firecrawl scrape https://example.com

# 2. Interact with it
firecrawl interact "Click the pricing tab"
firecrawl interact "Fill in the email field with [email protected]"
firecrawl interact "Extract the pricing table"

# 3. Code execution (Playwright)
firecrawl interact -c "await page.title()"
firecrawl interact -c "print(await page.title())" --python

# 4. Stop the session
firecrawl interact stop

Use --profile on the scrape to persist browser state across scrapes:

# Login and save state
firecrawl scrape "https://app.example.com/login" --profile my-app
firecrawl interact "Fill in email and click login"

# Come back authenticated later
firecrawl scrape "https://app.example.com/dashboard" --profile my-app
firecrawl interact "Extract the dashboard data"

Create monitors when the goal is ongoing change detection, alerting, or repeated checks over time. A monitor runs recurring scrapes or crawls, diffs each result against the last retained snapshot, and can notify webhooks or email recipients.

firecrawl monitor create --name "Blog" \
  --goal "Notify me when a new post is published" \
  --schedule "every 30 minutes" \
  --page https://example.com/blog \
  --email [email protected]

firecrawl monitor create --name "Product pages" \
  --goal "Notify me when pricing, docs, or changelog content changes" \
  --schedule "every 30 minutes" \
  --scrape-urls https://example.com/pricing,https://example.com/docs,https://example.com/changelog

firecrawl monitor create --name "Docs webhook" \
  --goal "Notify me when docs content changes" \
  --schedule "every 30 minutes" \
  --page https://example.com/docs \
  --webhook-url https://example.com/webhook \
  --webhook-events monitor.page,monitor.check.completed

firecrawl monitor list --limit 20
firecrawl monitor run <monitorId>
firecrawl monitor checks <monitorId>
firecrawl monitor check <monitorId> <checkId> --page-status changed
firecrawl monitor update <monitorId> --state paused
firecrawl monitor delete <monitorId>

For advanced targets such as JSON-mode changeTracking, pass a JSON payload:

firecrawl monitor create monitor.json
cat monitor.json | firecrawl monitor create

# Configure with custom API URL
firecrawl config --api-url https://firecrawl.mycompany.com
firecrawl config --api-url http://localhost:3002 --api-key fc-xxx

# View current configuration and authentication status
firecrawl view-config

Shows authentication status and stored credentials location.

# Login
firecrawl login
firecrawl login --method browser
firecrawl login --method manual
firecrawl login --api-key fc-xxx

# Login to self-hosted instance
firecrawl login --api-url https://firecrawl.mycompany.com
firecrawl login --api-url http://localhost:3002 --api-key fc-xxx

# Logout
firecrawl logout

These options work with any command:

firecrawl --status

  🔥 firecrawl cli v1.20.0

  ● Authenticated via stored credentials
  Concurrency: 0/100 jobs (parallel scrape limit)
  Credits: 500,000 / 1,000,000 (50% left this cycle)

# Output to stdout (default)
firecrawl https://example.com

# Pipe to another command
firecrawl https://example.com | head -50

# Save to file
firecrawl https://example.com -o output.md

# JSON output
firecrawl https://example.com --format links --pretty

• Single format: Outputs raw content (markdown text, HTML, etc.)
• Multiple formats: Outputs JSON with all requested data

# Raw markdown output
firecrawl https://example.com --format markdown

# JSON output with multiple formats
firecrawl https://example.com --format markdown,links,images

# Just pass multiple URLs -- results are saved to .firecrawl/
firecrawl scrape https://firecrawl.dev https://firecrawl.dev/blog https://docs.firecrawl.dev

# Many URLs, other formats, or JSON extraction: run one batch job
firecrawl batch-scrape --urls-file urls.txt --format markdown,links --wait -o results.json

# Extract links and process with jq
firecrawl https://example.com --format links | jq '.links[].url'

# Convert to PDF (with pandoc)
firecrawl https://example.com | pandoc -o document.pdf

# Search within scraped content
firecrawl https://example.com | grep -i "keyword"

# Set API key via environment
export FIRECRAWL_API_KEY=${{ secrets.FIRECRAWL_API_KEY }}
firecrawl crawl https://docs.example.com --wait -o docs.json

# Use self-hosted instance
export FIRECRAWL_API_URL=${{ secrets.FIRECRAWL_API_URL }}
firecrawl scrape https://example.com -o output.md

The CLI collects anonymous usage data during authentication to help improve the product:

• CLI version, OS, and Node.js version
• Detect development tools (e.g., Cursor, VS Code, Claude Code)

No command data, URLs, or file contents are collected via the CLI.

To disable telemetry, set the environment variable:

export FIRECRAWL_NO_TELEMETRY=1

Requests from the CLI are tagged integration: cli. A tool that wraps the CLI can tag them as its own by setting FIRECRAWL_INTEGRATION: the CLI sends the value, trimmed, as the X-Integration header on every API request (a value with control or non-ASCII characters cannot be a header and is not sent), and the API attributes the request to it when it is one of its integration names or any value starting with an underscore. Any other value is ignored by the API and never makes a request fail.

export FIRECRAWL_INTEGRATION=_my-tool

Supported API requests carry an origin of cli@<version>. When a coding agent runs the CLI, the origin also names the agent, for example cli.codex@<version>; it is detected from environment variables the agent sets in the commands it runs. Calls routed through the SDK include the agent only after the pinned SDK supports the origin option. To send plain cli@<version> instead:

export FIRECRAWL_CLI_DISABLE_AGENT_ORIGIN=1

Experimental commands live under firecrawl experimental (alias: firecrawl x).

Combines map + scrape to save a site as local files under .firecrawl/.

firecrawl x download https://docs.firecrawl.dev
firecrawl x download https://docs.firecrawl.dev --screenshot --limit 20 -y
firecrawl x download https://docs.firecrawl.dev --include-paths "/features,/sdks" -y

The old experimental AI workflow commands have moved to the firecrawl/skills catalog (skills/workflows/). Workflow skills infer from the user's request first and only ask short clarifying questions when required inputs are missing. Install them with:

firecrawl setup workflows

For more details, visit the Firecrawl Documentation.

Use a provider/capability in place of a URL. The explicit --alexandria form remains supported:

firecrawl scrape benzinga/news/search --options '{"pageSize":10}'
firecrawl scrape firecrawl-research-index/read --options '{"paperId":"123","query":"methodology","k":4}'
firecrawl scrape --alexandria benzinga/news/search --options '{"pageSize":10}'
firecrawl scrape https://example.com

URLs (including domains, IP addresses and localhost) continue to scrape websites. Tool addresses go directly to Alexandria, which validates the provider and capability; they never fall back to URL scraping. There is no extra catalog lookup. Bare names such as firecrawl scrape amazon fail locally with a suggested website URL and directions to firecrawl list. Suggestions are not verified or executed. Mixing URLs and tools in one command is rejected.

When a provider returns THIRD_PARTY_DATA_TERMS_REQUIRED, an organization admin must accept that provider's terms in the Firecrawl dashboard, at the link the error returns (requiresAction.url) or https://www.firecrawl.dev/app/settings?tab=data-sources. The CLI does not accept terms. To read the current provider agreement and metadata first:

npx firecrawl-cli@latest alexandria terms show benzinga --pretty

Agents must give the user the dashboard link and must not retry the provider until the terms are accepted. Reading terms does not accept them, and a failed lookup is never consent. After an admin accepts them, rerun the original provider command; its normal credits apply.

Alexandria execution JSON includes an additive receipt: creditsUsed is actual reported usage (missing means unknown), requestId is the client idempotency identity, and operationId/operationType identify the server scrape. Existing response fields remain available. IDs, reported credits, and available retry delays print to stderr.

Use the same request ID to recover pending or uncertain execution. Completed results, including failures, replay under the same ID; a deliberate new execution needs a new ID and may charge again. Never automatically rotate an uncertain ID. Structured failures preserve available status, code, action and retry metadata.

Report the outcome of a session, missing provider coverage, or capability issues:

firecrawl alexandria feedback --rating partial \
  --url https://example.com \
  --requested-functionality "Find records and download their attachments" \
  --objective "Compare contract requirements across agencies before bidding" \
  --rationale "Found summaries but could not retrieve attachments" --json

The optional --objective is the underlying goal of the session: what you or your user were ultimately trying to accomplish, beyond the single website. No job ID is required. It must be sent within 20 minutes of the team's most recent Alexandria search, discovery, or execution; later submissions are rejected with FEEDBACK_WINDOW_EXPIRED. Each submission refunds 1 credit, up to 10 credits per website and 100 credits per team each UTC day; past either cap, submissions are still recorded and return websiteCapReached or dailyCapReached. Optional --provider-feedback and --capability-feedback accept JSON arrays; see firecrawl alexandria feedback --help for their fields and issue codes. Capability issue codes are new_capability_request (requires requestedFunctionality), missing_capability (the provider exists but lacks this capability), insufficient_functionality, incorrect_result, execution_error, and other. Each eligible provider call prints a Feedback ref (for example c3) so ratings are attributed to the calls they are about. The CLI keeps up to 20 call tokens for 24 hours in its config directory and sends them with the session rating, then clears them after a successful submission. Rate individual calls with a repeatable --call-rating c3=good, or pass tokens yourself with a repeatable --token <token> (these replace the stored ones). Existing feedback and search-feedback commands retain their job-specific behavior. Endpoint feedback opt-out environment variables also apply to this command.