Firecrawl API for AI agents: scrape, crawl and map sites without a key

    Give Claude, ChatGPT, Cursor or Codex the Firecrawl API: turn a page into clean Markdown, map a site's URLs, crawl a section, scrape a batch of pages, search the web or let Firecrawl's agent find data across a site. You don't need a Firecrawl account or key. Your agent pays per use from your Agensi balance.

    6 tools on Agensi
    Data 6
    firecrawl.dev

    Use any of these from your AI Agent, with no Firecrawl account or API key. Your AI Agent works out the exact price, runs the tool from your balance and tells you what it cost. Failed uses are free.

    Connect your AI Agent

    All Firecrawl tools and prices

    • Crawl a website

      $0.065 per 10 pages

      Data
      Proven · 100% success

      Start a bounded crawl and pay for pages processed.

    • Extract a batch of pages

      $0.065 per page

      Data
      Proven · 100% success

      Start a bounded batch scrape of up to ten URLs.

    • Map a website

      $0.0065 per call

      Data
      Proven · 100% success

      Discover URLs on a website without scraping their content.

    • Search the open web

      $0.013 per call

      Data
      Proven · 100% success

      Find up to ten web results without scraping their full pages.

    • Firecrawl Agent

      $0.0013 per call

      Data
      New

      Describe the data you want in natural language and an autonomous agent browses, navigates, and extracts it into JSON — optionally shaped by your 'schema' and constrained to given 'urls'. Use it when the pages holding the answer are unknown or spread across a site; use a plain scrape when you already know the one URL. Pricing is dynamic with a hard ceiling you set via 'maxCredits' (default 100 here): the run settles at what the agent's research actually consumed, never above the ceiling. Runs take minutes and are polled internally until the extraction completes.

    • Firecrawl Scrape

      $0.0013 per call

      Data
      New

      Turn a single web page URL into clean, LLM-ready Markdown: the main content without navigation, ads, or boilerplate, ready for prompts, RAG, and analysis. Also returns HTML, links, images, or a summary. JavaScript-rendered pages and anti-bot blocks are handled at the same price; PDFs parse natively and bill per PDF page. Set 'json' to true for structured JSON extracted by an LLM, at five times a plain scrape. A page answering 403 or 404 is still charged — check metadata.statusCode before retrying a blocked URL.

    Starting prices. Prices can vary with length and quality. Your AI Agent works out the exact price, runs the tool from your balance and tells you what it cost. Failed uses are free.

    What do you want your agent to do?

    • Read one page

      Clean Markdown of a page's main content without menus and ads. JavaScript pages and PDFs work too.

    • List a site's pages

      The URLs on a site, found without scraping their content.

    • Crawl a section

      A bounded crawl of a site or section, charged for the pages it processes.

    • Scrape a list of URLs

      Up to ten known pages in one batch.

    • Search the web

      Up to ten results for a query, without scraping the full pages.

    • Let an agent find the data

      Describe what you want and Firecrawl's agent browses the site and returns JSON. Runs take minutes.

    How to choose the best Firecrawl tool or MCP for you

    Do you know the URL?

    If you know the page, scrape it. If you know the site, map it first and scrape only the pages you need. If you don't know where the data is, use the agent.

    Size

    Crawls are charged for the pages they process, so map the site first and start with a small crawl before you go wide.

    Output

    Scrape returns Markdown and can also return HTML, links or a summary. Ask for JSON when you want fields extracted; that costs more per page.

    Error pages

    A page that answers with an error such as 404 still counts as a scrape, so have your agent check the status before it retries a URL.

    If you needChooseToday's price
    The text of one pageFirecrawl Scrape
    The URLs on a siteMap a website
    A docs or blog sectionMap, then crawl that section
    Ten known URLs at onceBatch scrape
    Pages about a topicSearch the open web
    Data spread over a site you don't knowFirecrawl Agent

    Quick picks: scrape for one page, map before any crawl, the agent when you don't know where the data is.

    Firecrawl's own MCP server or Agensi?

    Firecrawl publishes its own MCP server. It needs a Firecrawl API key for most tools and uses the credits on your Firecrawl plan. With Agensi your agent uses Firecrawl's scrape, map, crawl, batch scrape, search and agent without a Firecrawl account, on the same connection as search, social and company data tools, paid per use from one balance. Your agent runs a tool from your balance and tells you what it made and what it cost. Failed uses are free.

    What you getOfficial MCP per brandOwn API key or community serverAgensi
    AccountsOne per brandOne per brandOne
    API keysPer brand, or a sign-inPer brand, kept by youNone
    PaymentA plan or credits per brandPer brandOne balance, per use
    Spending controlVariesUp to youLimits per use, day and month, or ask first
    Several sources in one answerConnect several serversSeveral servers and keysOne connection
    Skills for the whole jobNoNoYes, from the marketplace

    When another way is the better choice. Firecrawl's own MCP server is the better pick if you crawl at high volume on a Firecrawl plan, or need tools we don't list, such as document parsing or its people and company enrichment. Its hosted server also lets you try scrape, search and parse without a key, with rate limits.

    Set up in your AI agent

    Connect your AI agent to Agensi once, sign in and click Allow. After that it can use every tool: it runs one from your balance and tells you what it cost. No API keys or provider accounts.

    1. Open Settings, then Connectors, then Browse connectors and add Agensi (or Add custom connector and paste https://mcp.agensi.io/mcp/auth).
    2. Sign in to Agensi and click Allow. One agent per account: connecting a new one disconnects the old one.
    3. Ask for what you want, for example: "Map our competitor's website, scrape their pricing page and three product pages, and compare their plans with ours." Top up from $5 if your balance is low; failed uses are free.

    Grok, VS Code, GitHub Copilot, Cline, Kiro and other agents with remote MCP and sign-in work too: see the setup guide.

    What it costs: a worked example

    "Map our competitor's website, scrape their pricing page and three product pages, and compare their plans with ours."

    • 1 × Map the site's URLs

    • 4 × Scrape four pages to Markdown

    Your agent works out the exact price for your request and tells you what it cost.

    Combine Firecrawl tools with skills

    Firecrawl MCP: frequently asked questions

    The tools run on Firecrawl's API through our account; Agensi isn't made by Firecrawl. Firecrawl also publishes its own MCP server, which uses your Firecrawl key and credits. With Agensi you skip the key and pay per use from the same balance you use for every other tool.

    Use it on public pages that a site allows to be read. Respect robots.txt and each site's terms, don't collect personal data without a legal basis, and keep the volume of requests reasonable. This isn't legal advice, so check the rules that apply to you and your use.

    Each tool shows its live price on this page. A scrape is charged per page, with more for JSON extraction, and a crawl for the pages it processes. Your agent works out the exact price, runs the tool from your balance and tells you what it cost. Failed uses are free.

    Yes. Connect Agensi once in Claude Code, Cursor, Codex, Claude or ChatGPT; the setup steps are on this page. In Claude Code and Cursor your agent can save the Markdown it scrapes straight into your project, for docs, tests or a knowledge base.

    Read next

    Browse all tools