firecrawl-agent Skill: AI Extraction of Structured Data From Websites

Makes your coding assistant run Firecrawl's AI agent, which navigates sites and extracts structured data across pages.

Installs: 81.6K; on skills.sh
Stars: 676; on GitHub
Updated: Oct 2026; last repo commit

firecrawl-agent teaches your assistant to run firecrawl agent "<prompt>". Firecrawl's agent navigates a site on its own and returns structured data, and a run takes 2 to 5 minutes. You can give it a JSON schema for predictable output or a list of URLs to focus on.

You must install the Firecrawl CLI and sign in, because the agent has no free keyless tier. Agent runs use more credits than a simple scrape, and --max-credits caps the spend.

Good for

Developers who need pricing tiers, product lists or features pulled from complex sites
Sales and research teams who want contact or company data from ready-made data providers
Anyone whose data sits behind several pages of navigation

Install firecrawl-agent

Claude Code

Copy this prompt.

Install the firecrawl-agent skill in Claude Code. Run this command:

npm install -g firecrawl-cli@latest
firecrawl login --browser
npx skills add https://github.com/firecrawl/cli --skill firecrawl-agent -a claude-code

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js, A Firecrawl account (or the keyless free tier).

The skill only gives your assistant instructions. The Firecrawl command line tool does the work, so install it and sign in first. Agent needs an account: the install rules list it among the commands that prompt a login when no credentials are set. The one-command route installs the CLI, signs you in through the browser and adds the core and workflow skills for every editor it detects, not just this skill.

From the vendor's docs. Source.

What's in the skill

When it loads

Autonomously navigate websites and extract structured data across pages. Use when the task requires navigation or no suitable ready-made workflow or data provider covers it.

Sections in SKILL.md

Quick start
Alexandria providers
Job IDs
Tips
See also
Alexandria session feedback

Key facts

Author: firecrawl org
Repo: firecrawl/cli
Installs on skills.sh: 81,646 as of 2026-11-02
GitHub stars: 676 for the whole firecrawl/cli repo, 2026-11-02
License: ISC (package.json of the CLI repo)
Last commit to the repo: 2026-10-28
What is inside: 1 file (SKILL.md)
Needs: The Firecrawl CLI (npm package firecrawl-cli), Node.js and a Firecrawl account. The install rules list agent among the commands that prompt a login when no credentials are set.
Related skills: firecrawl-scrape, firecrawl-interact, firecrawl-crawl, firecrawl-alexandria
skills.sh audits: Gen Agent Trust Hub pass, Socket pass, Snyk warn (2026-11-02)

What it includes

Extracts structured data by navigating a site
Takes 2 to 5 minutes. --wait returns the result inline, and without it you get a job ID to check later.
Try: “Use the Firecrawl agent to extract all pricing tiers from example.com/pricing.”
Runs code · firecrawl agent "<prompt>" --wait --json -o
Fixes the shape of the output with a JSON schema
Without a schema the agent returns freeform data.
Network · --schema
Focuses the agent on specific pages
Example in the skill: get a feature list from one URL.
Network · --urls
Caps how many credits a run can use
The skill says agent runs use more credits than simple scrapes.
Network · --max-credits
Checks, waits on or cancels a running job
A job ID as the prompt checks status. Add --wait, --poll-interval and --timeout to wait, or --cancel to stop it.
Runs code · Job IDs: status, wait, cancel
Lets a run call connected data providers
Up to 5 providers per run. A run uses providers only when it starts with one of the Alexandria flags.
Try: “Find the head of sales at Acme with the apollo and crunchbase toolkits.”
Runs code · --toolkits, --max-calls, --alexandria
Asks you before a paid provider call
Needs --mode chat. The run can stop on a pending approval, and the assistant answers it on the same thread. Terms approvals need an organization admin to accept in the dashboard.
Runs code · --require-approval, --approve, --decline
Reports a session's outcome or a coverage gap
Sends a rating, a website, the requested functionality, an objective and a rationale. Each submission refunds 1 credit, within 20 minutes of the last Alexandria call.
Runs code · firecrawl alexandria feedback

From the official docs

Permissions

Medium risk

This skill sends your prompts to Firecrawl's agent, which uses more credits than a scrape and can call paid data providers. Set a credit cap and ask for approval before paid calls.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Agent runs use more credits than a scrape
The skill says so and offers --max-credits as the cap. Always set a cap, and prefer scrape for a single page. github.com
Data-provider calls can be paid
Runs that start with --toolkits or --alexandria can call connected providers, up to 5. Use --require-approval so the run stops before a paid call. github.com
It can send session feedback to Firecrawl
The alexandria feedback command sends a rating, a website, your objective and a rationale. Check what the assistant writes in it. github.com
Web pages can try to steer your assistant
The skill's own security note calls fetched web content untrusted and says it may contain prompt injection. Do not let the assistant follow instructions it finds on a page. github.com
Setup installs a global npm package
The CLI is installed with npm install -g. The one-command setup also adds skills to every coding editor it detects, so use the single-skill route if you want only this one. github.com
The CLI sends anonymous usage data
The README says it collects the CLI version, OS, Node.js version and which dev tools it detects during sign-in. Set FIRECRAWL_NO_TELEMETRY=1 to turn it off. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	No read-only mode. The skill runs CLI commands that spend credits and write files.
Talks to	Firecrawl's API, or your own server if you set --api-url. Web pages the CLI fetches. Page content the assistant reads goes to your AI provider as part of the chat
Sign-in and keys	Browser sign-in stores credentials in the CLI. You can set FIRECRAWL_API_KEY instead.
Scopes	Runs firecrawl commands in your shell (the skill allows Bash(firecrawl ) and Bash(npx firecrawl-cli ))

Questions

What does firecrawl-agent do?

It makes your assistant run Firecrawl's agent, which navigates websites on its own and returns structured data. You can give it a prompt, a JSON schema and URLs.

How do I install firecrawl-agent?

Use the install box at the top of this page. You need the Firecrawl CLI and a sign-in as well as the skill. The one-command route installs the whole core skill set.

Do I need a Firecrawl account?

Yes. The install rules list agent among the commands that prompt a login when no credentials are set. Sign in with the browser or an API key.

How much does an agent run cost?

The SKILL.md only says agent runs use more credits than simple scrapes, and that --max-credits caps the spend. It gives no price. For one page, the skill says to use scrape instead because it is faster and cheaper.

Is firecrawl-agent safe to use?

It sends your prompt and URLs to Firecrawl and can spend credits, including on paid data providers. Use --max-credits, and add --require-approval if you want a stop before paid provider calls.

firecrawl-agent or firecrawl-interact?

The SKILL.md describes interact as manual control of a page and agent as AI-powered extraction with less control. For plain bulk extraction without AI it points to firecrawl-crawl.

Skill or Firecrawl MCP server?

They are two routes to Firecrawl. firecrawl-agent is a SKILL.md that tells the assistant to run the Firecrawl CLI in your shell. The skill only allows Bash(firecrawl *) and Bash(npx firecrawl-cli *), so it needs an assistant that can run shell commands. The Firecrawl MCP server and the local Firecrawl MCP server give your client Firecrawl as tools it calls directly, with no skill involved. The README describes firecrawl setup mcp for adding the MCP server to your editors. You can use either route, or both.

Information

updated Nov 2, 2026

Repository: firecrawl/cli
Folder: skills/firecrawl-agent

firecrawl/cli skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 5

Firecrawl CLI install rules · github.com
firecrawl/cli README · github.com
skills CLI commands · github.com
firecrawl-agent SKILL.md · github.com
Firecrawl security rules · github.com