Firecrawl Action

Scrape, crawl, search, and map web pages as part of your webhook pipelines. Extract structured data, convert pages to markdown, discover URLs, or search the web — all powered by the Firecrawl API. You provide your own Firecrawl API key.

Overview

The Firecrawl Action node connects to the Firecrawl API to perform web scraping operations. It supports four modes — picked under Operation on the node page — each suited for different use cases, or all four as an AI toolkit. The node appears on the canvas with an orange color.

A new Firecrawl node starts with no operation picked, and it doesn't run until you pick one — so nothing gets scraped, or charged to your Firecrawl credits, by a node nobody set up.

Modes

Scrape

Extracts content from a single URL. Returns markdown, HTML, screenshots, or structured JSON. Best for extracting data from a specific page — product prices, article content, form data, etc.

Crawl

Recursively crawls an entire website starting from a base URL. Discovers and scrapes multiple pages following links. Returns an array of scraped pages — the output is marked as iterable so downstream nodes process each page individually.

By default a crawl only follows pages under the start URL's path: starting at https://site.com/docs/intro it reads/docs/intro and what hangs below it, and starting at a single article it reads just that article. Turn on Crawl the whole site (v2) to also follow links to sibling and parent pages of the same site.

Search

Searches the web for a query and returns scraped content from the results. Combines web search with page scraping in one step. Returns an array of results — also iterable.

Map

Discovers all URLs on a website without scraping them. Returns a list of links with titles and descriptions. Useful for discovering pages before selectively scraping them with a downstream Firecrawl node. Output is iterable.

Node versions

Firecrawl is the first node type with two versions. The pill next to the breadcrumb on the node's page (v1 / v2) says which one a node is on and opens the picker to change it.

Switching a node to another version never touches its name, its connections or whether it is active. Going from v1 to v2 loses nothing. Going back to v1 removes the five v2-only settings (onlyCleanContent, removeBase64Images, location, maxAge, scrapeResults) — the dialog lists exactly what goes — and the change can be undone from the same pill.

Configuration

Common Fields

Scrape / Crawl Fields

Crawl-Specific Fields

Search-Specific Fields

Map-Specific Fields

Structured Extraction (Scrape / Search)

Ask Firecrawl to extract structured data from the page with a prompt, a schema, or both. It is sent as the json format and the result lands in json in the output. On v2, turn on the JSON format chip and the section appears; turn it off and no extraction is requested (your prompt and schema are kept for later). On v1 only the prompt is available.

The schema box takes one field per line, field: type. This asks for a name, a price, a stock flag, a list of tags, a nested author and a free-text summary:

name: string
price: number
inStock: boolean
tags: string[]
author.name: string
summary: A one-line summary of the page

Types are string, number, integer, boolean, object, array, and any of them with [] for a list. Dots nest (author.name). Any other text is read as the description of a text field — a good place to tell the model what you mean. Every field you list is requested. The node turns this into a JSON Schema when it calls Firecrawl; a full JSON Schema pasted as-is works too.

{
  "json": {
    "name": "Nest: Desarrollo backend escalable con Node",
    "price": 19.99,
    "inStock": true,
    "tags": ["backend", "typescript"],
    "author": { "name": "Fernando Herrera" },
    "summary": "A hands-on course on building APIs with NestJS."
  },
  "metadata": { "title": "...", "url": "...", "statusCode": 200 }
}

v2-only fields

Output Payload

Scrape Output

Returns a single object with the scraped content. Not iterable.

{
  "_meta": { "iterable": false, "count": 1 },
  "markdown": "# Page Title\n\nPage content...",
  "links": ["https://example.com/about", "https://example.com/pricing"],
  "json": { "price": 19.99, "inStock": true },
  "metadata": {
    "title": "Page Title",
    "url": "https://example.com",
    "statusCode": 200
  }
}

Every format you asked for arrives as its own field — html, rawHtml, images, summary, screenshot (a URL that expires in 24 hours) and so on.

Crawl Output

Returns an array of pages. Marked as iterable — downstream nodes process each page individually.

{
  "_meta": { "iterable": true, "iterateField": "pages", "count": 5 },
  "pages": [
    { "markdown": "# Page 1...", "metadata": { "sourceURL": "https://example.com/" } },
    { "markdown": "# Page 2...", "metadata": { "sourceURL": "https://example.com/about" } }
  ]
}

Search Output

Returns an array of search results with scraped content. iterable.

{
  "_meta": { "iterable": true, "iterateField": "results", "count": 5 },
  "results": [
    { "url": "https://example.com", "title": "Example", "markdown": "..." },
    { "url": "https://other.com", "title": "Other", "markdown": "..." }
  ]
}

Map Output

Returns an array of discovered URLs. iterable.

{
  "_meta": { "iterable": true, "iterateField": "links", "count": 150 },
  "links": [
    { "url": "https://example.com/", "title": "Home", "description": "..." },
    { "url": "https://example.com/about", "title": "About Us" }
  ]
}

AI toolkit mode

Once the API key is picked, the node asks for a Mode. Single operation is everything above: one of the four operations, your templates, every run. AI toolkit turns the node into four tools that an AI Node — or an MCP client, through one of your MCP servers — calls by name, with the URL or the query and every option of the v2 decided per call.

Lists go comma-separated. formats takes markdown, html, rawHtml, links, images, summary, screenshot and json (with extract_prompt or extract_schema); sources takes web, news and images. The tools always run as the v2, whatever version the node is on.

What you control

Very little, on purpose: the model decides the rest from what the user asks. What you set in Single operation — formats, mobile, country, limits — does not apply to the tools. The two settings live in the Settings card of the node's AI Toolkit section, and both are optional.

• Allowed domains. With a list, the model can only scrape, crawl and map those domains and their subdomains, a crawl never follows links out of them, and a search only looks inside them — the query is limited with site:, so no credits go to reading results that would be thrown away. Anything from another domain that still comes back is dropped before the model sees it. Empty means any site.
• Characters per page sent to the model. 20,000 by default (about 5,000 tokens), anything from 1,000 to 200,000. Lower it to save tokens; raise it when the agent needs whole long pages and its model has the context for them.
• Fixed limits. Not a setting: a crawl reads at most 50 pages and waits between 30 seconds and 5 minutes, a search returns at most 20 results and a map at most 5,000 URLs. When the model asks for more (or less), the call runs at the limit and the answer says so. Without instructions it uses Firecrawl's usual defaults — markdown, main content only, ads blocked, 10 pages, 5 results — and a map returns 100 URLs. crawl_site only follows pages under its start URL unless the model passes crawl_entire_domain.
• Crawling starts blocked. One crawl_site call can spend many credits and take minutes. Unblock it under Block these tool calls if the agent needs it; block any other tool the same way.

What goes back to the model

The same response Firecrawl returns, with two cuts so it fits a model's context: each page's markdown, HTML and summary stop at the node's characters-per-page limit (20,000 unless you change it), and link and image lists at 500. A map's list of URLs is cut to fit the same characters limit. Every cut is listed in a notes field, so the model knows it didn't get everything. Arguments are used as written: a {{payload.x}} the model writes is not resolved, so it can't read the output of other nodes in the flow.

Testing a tool

{ "operation": "scrape_url", "url": "https://example.com/pricing", "formats": "markdown,links" }

Use Cases

Price Tracking

Webhook (product URL) --> Firecrawl (scrape) --> Transform (extract price) --> Google Sheets (save)

Competitor Monitoring

Scheduled Workflow (daily) --> Firecrawl (crawl competitor site) --> Filter (changed pages) --> Email (report)

Web Search + AI Summary

Webhook (search query) --> Firecrawl (search) --> AI Node (summarize) --> Notification (Discord)

Site Discovery + Selective Scraping

Webhook (URL) --> Firecrawl (map) --> Split (per URL) --> Firecrawl (scrape each) --> MongoDB (save)

Research agent (AI toolkit)

Chat Trigger --> AI Node  ·  tools: Firecrawl toolkit (search_web, map_site, scrape_url)

The agent searches, maps the site it lands on and reads the pages that answer the question — each step decided per turn. Add your own domains under Allowed domains to make it a support agent that only reads your documentation.

API Key Setup

The Firecrawl node requires a Firecrawl API key. Get one at firecrawl.dev. In the node's detail page, click "Add Firecrawl API Key" to store it securely in the credentials system. The key is encrypted at rest and never exposed in API responses.

API Reference

Create

curl -X POST /api/firecrawl-actions \
  -H "Authorization: Bearer hwk_..." \
  -d '{
    "workspaceId": "<your-workspace-id>",
    "name": "Scrape Product",
    "mode": "scrape",
    "url": "{{payload.url}}",
    "formats": ["markdown"],
    "onlyMainContent": true,
    "credentialId": "..."
  }'

Test

curl -X POST /api/firecrawl-actions/:id/test \
  -H "Authorization: Bearer hwk_..." \
  -d '{ "payload": { "url": "https://example.com" } }'