Summary: Firecrawl is a web API that turns pages into clean Markdown or structured JSON that you can feed to an LLM, a RAG index or an AI agent. This tutorial walks through the three calls most people actually need (scrape one page, map a site's URLs, crawl a section), how credits are counted, and the settings that stop a first crawl from burning your whole free allowance. Everything below was checked against Firecrawl's own docs and pricing page on 2026-11-03.
What Firecrawl does, in plain terms
Feeding raw HTML to a model wastes tokens on menus, cookie banners and scripts. Firecrawl fetches the page for you (including JavaScript-rendered pages), strips the clutter and hands back the content in the format you ask for. The docs list more than a dozen output formats; the ones you will use first are markdown, html (cleaned), links, screenshot and json for structured extraction with a schema.
There are four core building blocks worth knowing before you write any code:
- Scrape: one URL in, clean content out. 1 credit per page.
- Map: give it a site, get back a list of URLs, mostly from the sitemap plus search results. 1 credit per call, no matter how many URLs come back.
- Crawl: start from a URL and scrape every reachable subpage. 1 credit per page crawled.
- Search: run a web search and optionally scrape the results. 2 credits per 10 results, plus scrape costs for any result you scrape.
Step 1: Decide between keyless access and a free API key
Firecrawl now lets you call Scrape, Search and Interact with no API key at all. That keyless tier is free but capped per IP address per day, by both a request count and a credit count, and going over either returns HTTP 429. Crawl, Map, Extract and batch scrape are not available without a key.
One thing the docs do not spell out up front: keyless access depends on where you call from. When I sent the documented keyless scrape request from a cloud server while writing this post, the API answered with HTTP 403 and a message saying the IP address looked like a VPN or proxy, so it could not be used without an API key. If you work from a VPS, a CI runner or a corporate VPN, plan on getting a key.
The free account is the better starting point anyway. It costs nothing, needs no card, and gives you 1,000 credits per month (they reset monthly and do not roll over), 2 concurrent browsers, and access to every core endpoint. To get one:
- Sign up at firecrawl.dev and open the API keys page in the dashboard.
- Copy the key (it starts with
fc-) and store it as an environment variable namedFIRECRAWL_API_KEYrather than pasting it into code. - Optional but smart: open the menu on the key, click Set limit, and set a credit cap per day, week or month. The limit is counted in credits, not dollars, and applies to that key alone.
Step 2: Scrape your first page
The quickest test is a single cURL call to the v2 scrape endpoint:
curl -s -X POST "https://api.firecrawl.dev/v2/scrape" \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $FIRECRAWL_API_KEY" \ -d '{"url": "https://example.com", "formats": ["markdown"]}'
In Python, install the SDK with pip install firecrawl and the same call becomes:
from firecrawl import Firecrawl
fc = Firecrawl(api_key="fc-YOUR-API-KEY")
doc = fc.scrape("https://example.com", formats=["markdown", "links"])
print(doc)
Read the response correctly. Firecrawl has two status layers. HTTP 200 with success: true only means your API call worked. The target site's own answer is in data.metadata.statusCode. A page that returned 403 or 404 still comes back as a document, and per the billing docs it still costs 1 credit. Check that field and stop retrying URLs that keep failing, or a loop will quietly drain your balance.
Step 3: Map the site before you crawl it
This is the habit that saves the most credits. A map call costs 1 credit even if it returns thousands of URLs, so it is the cheapest way to see how big a site really is and which paths you care about.
res = fc.map(url="https://docs.example.com", limit=50, sitemap="include")
Map also takes a search parameter to return only URLs related to a topic, which is handy when you only want, say, the pricing or API pages of a large site. Once you have the list, you can either scrape just those URLs or design a tight crawl.
Step 4: Run a crawl with a hard limit
Crawl follows links from your starting URL and scrapes each page it finds. The default page limit is 10,000, which is far more than a free account has, so always pass your own:
docs = fc.crawl(
url="https://docs.example.com/guides/",
limit=100,
scrape_options={"formats": ["markdown"], "only_main_content": True}
)
Settings that matter, all from the crawl reference:
- Children only by default. Starting at
/guides/will not reach/blog/. UsecrawlEntireDomainfor sibling and parent paths andallowSubdomainsfor subdomains. includePathsandexcludePathsare regex patterns matched against the URL path, not the full URL. If your starting URL does not matchincludePaths, the crawl may return 0 pages.maxDiscoveryDepthcounts link hops from the root, not slashes in the URL.- Credits arrive late. Crawl and batch jobs are billed as each page finishes, so your dashboard can keep ticking up for minutes after you submit. Polling job status is free.
Every scrape option (formats, proxy, caching, actions, location) also works inside a crawl and applies to every page, which means a costly option multiplies fast. JSON extraction adds 4 credits per page, so a 100-page crawl with a JSON schema costs about 500 credits instead of 100.
Step 5 (optional): Plug Firecrawl into your coding agent
If you mainly want your AI assistant to read live pages, you may not need to write code. Firecrawl runs a hosted MCP server; its keyless URL exposes Search, Scrape and Parse. The docs give one-line setups, for example claude mcp add --transport http firecrawl https://mcp.firecrawl.dev/v2/mcp for Claude Code and codex mcp add firecrawl --url https://mcp.firecrawl.dev/v2/mcp for Codex CLI. There is also a CLI installer, npx -y firecrawl-cli@latest init --all --browser, that signs you in through the browser and adds skills to detected coding agents; restart the agent afterwards so it picks them up.
Free vs Hobby vs Standard
Prices are from firecrawl.dev/pricing in USD; plan limits are from Firecrawl's billing and rate-limit docs.
| Plan | Price | Credits per month | Concurrent browsers | /scrape and /crawl per minute | When credits run out |
|---|---|---|---|---|---|
| Free | $0, no card | 1,000 (about 1,000 pages or 500 searches) | 2 | 10 and 2 | HTTP 402 until the monthly reset; no pay-as-you-go |
| Hobby | $19/month, or $16/month billed yearly | 5,000 | 5 | 100 and 20 | Optional pay-as-you-go: 1,000 credits per $5 |
| Standard | $99/month, or $83/month billed yearly | 100,000 | 25 | 500 and 100 | Optional pay-as-you-go: 2,000 credits per $5 |
Bigger Growth ($399/month, 500,000 credits) and Scale ($749/month, 1,000,000 credits) plans exist for high volume. Note that upgrades are charged at the full new price on the day you upgrade, with no proration, and your billing cycle restarts.
Editor's note
Good fit: developers building RAG pipelines, agents that need live web context, or anyone who wants a docs site as Markdown without maintaining their own headless browser and proxy setup.
Not a fit: one-off reading of a single article (your AI chat app's browsing is enough), or teams who need guaranteed fixed monthly costs but leave pay-as-you-go uncapped.
Gotchas I verified:
- 403 on keyless calls from servers: the API rejected my keyless request because the IP looked like a VPN or proxy. Fix: add an API key.
- 402 Payment Required: you are out of credits, or a crawl's
limitcould not fit your balance. Fix: lowerlimit, wait for the reset, or turn on pay-as-you-go on a paid plan. - 429 Rate limit exceeded / Concurrency limit reached: back off and retry after the
Retry-Afterheader. Free accounts hit this quickly on crawl (2 requests per minute). - 401 Unauthorized: the key is missing or malformed; the header must be
Authorization: Bearer fc-.... - 400 on crawl filters: path patterns use Rust regex, so look-arounds and backreferences are rejected rather than ignored.
- Screenshot URLs expire after 24 hours, so download them if you need to keep them.
FAQ
Is Firecrawl free?
There is a genuinely free plan with 1,000 credits per month and no card, plus a keyless tier for Scrape, Search and Interact that is capped per IP per day. Heavy or repeated crawling needs a paid plan.
Does a failed page cost credits?
If Firecrawl returns a document, you pay 1 credit even when the target site answered 403 or 404. If Firecrawl returns no document at all, for example because the site never responded, the scrape is not charged.
Should I use Map or Crawl?
Use Map first to see the URL list for 1 credit, then crawl only the section you need with a limit and path filters. Crawl without a plan is the fastest way to spend a monthly allowance.
Related guides on this blog
- Dify Playbook: Turn Your Help Docs Into a Support Chatbot (Dify's website import uses Firecrawl as a data source)
- Pinecone vs Qdrant vs Weaviate for Your First RAG App (where your scraped Markdown goes next)
- Tavily vs Exa vs Brave Search API: Free Tiers and Real Costs (if you need search results more than full pages)
Map first, crawl with a limit, cap your key, and Firecrawl's free 1,000 credits go a long way.
Images: own screenshots of Firecrawl's official pricing page, documentation and playground (firecrawl.dev, docs.firecrawl.dev); see them at the end of this post.
Last verified: 2026-11-03
Sources:
Firecrawl Docs: Introduction
Firecrawl Docs: Scrape
Firecrawl Docs: Map
Firecrawl Docs: Crawl
Firecrawl Docs: Billing
Firecrawl Docs: Rate Limits
Firecrawl Docs: Errors
Firecrawl Pricing
댓글
댓글 쓰기