firecrawl
The front door to Firecrawl's web tools: say what you want off the internet in plain words, and it picks the right tool for the job and runs it.
Included with installation: 13 skills
Skills
This plugin's skills 1313 of 13
SETUP
Showing all 13 skills
Finding the page3 skills
For when the thing you want is somewhere on the internet and you do not know the address. One of these three is the catch-all entry point: given a request in plain words it decides which of the rest to start, so an assistant that has been told nothing specific begins here.
Open skill
firecrawl-map
You know the site but not which page on it — this pulls up the full list of its page addresses and lets you search that list for the one you want.
firecrawl-search
Search the web by asking your real question, and get back not just links and excerpts but, if you want, the full text of every page it found.
Reading the page3 skills
Once an address is in hand, these turn it into words: the one page you named, every page under a section of a site, or a named set of fields filled in from however many pages that takes.
firecrawl-agent
Set it loose on a website and a few minutes later it comes back with the specific facts you asked for, gathered from however many pages it had to visit.
firecrawl-crawl
Instead of opening a site's pages one by one, this collects a whole section in a single run and hands back all the text together.
firecrawl-scrape
Give it a web address and it hands back that page's text, tidied up and saved to a file — including pages that only fill themselves in once a browser loads them.
Pages that need working2 skills
Two cases a plain fetch cannot settle. A site that shows nothing until somebody logs in, clicks or scrolls through to the next lot. And a site whose value is in how it changes, so something has to keep looking and say when it does.
firecrawl-interact
Some pages give up nothing until someone clicks. This one works the page the way a person would — logging in, filling in a form, moving to the next page — until what you wanted is on screen.
firecrawl-monitor
Keeps checking a page, a whole site, or even the open web on a schedule you set, and emails you the moment something you care about changes or shows up.
Files on your disk2 skills
These two finish with files rather than with words in the conversation: a whole site saved for reading with no internet, and a document you already have — a PDF, a spreadsheet — turned into something plain.
firecrawl-download
A copy of a website on your own computer: every page saved as a file you can open when you're offline, with screenshots if you ask for them.
firecrawl-parse
Takes a document already sitting on your computer — a PDF, a Word file, a spreadsheet — and turns it into plain readable text, or just answers a question about what's inside.
Indexes Firecrawl keeps3 skills
Three ways round the open web: a directory of ready-made workflows and data sources, an index of public code, issues and documentation for programming questions, and a paper index weighted to biomedical and life-science research.
firecrawl-alexandria
Before reading a website page by page, this checks Firecrawl's catalogue of ready-made data sources for one that already covers that site, and gets the information back as tidy, organised data instead.
firecrawl-developer-index
Answers programming questions from the original source — the bug report where a problem was filed, the merged change that fixed it, the official page that says how something works — rather than from a blog post describing it.
firecrawl-research-index
Finds the research papers that answer a question — mostly medical and life-science journals and preprints, plus computer science, physics and maths — and brings back the neighbouring papers around the best match, not just one.
No skills in this package match the current search or filters.
Before you install
An account, and a switch that changes your editor.
Every command goes through a Firecrawl account: sign in through the browser the first time, or put a key in FIRECRAWL_API_KEY. Pointed at a Firecrawl you run yourself with --api-url, it skips the sign-in entirely. Every command spends credits from that account; an agent job can be capped with --max-credits, and fails rather than go over.
One setup step changes settings rather than adding files:
- firecrawl setup defaults turns off the assistant's own web fetch and search, where the assistant allows that
- run on its own it asks one assistant at a time, so you can change some and leave others alone
- the same command with --undo puts the original settings back
INSTALL
One line does the whole setup:
it signs you in to Firecrawl in the browser, installs the firecrawl command and adds Firecrawl's CLI and workflow skills to every coding agent it finds.
Run this command in a terminal
npx -y firecrawl-cli@latest init --all --browser