Firecrawl
AIGive agents production-grade web data through Firecrawl. They scrape single URLs into clean Markdown, crawl entire sites, map site structure, search the web, and extract structured JSON with LLM-powered extraction, with credit and token usage tracking built in.
What This Integration Enables
Firecrawl turns the open web into agent-ready data: any page as clean Markdown, any site section as structured JSON extracted against a prompt or schema, any topic as searched-and-scraped results in a single call. FlowRunner agents use it as their eyes on the web. They scrape single URLs synchronously, run batch scrapes and full crawls as tracked jobs, map a site's URL inventory in seconds before deciding what to fetch, and run LLM extraction across glob patterns like an entire blog or catalog section. The discipline this connector rewards is treating web data as a metered input, because it is one. Every scrape and crawl spends from a finite credit pool, extraction spends tokens, and Cancel Crawl does not refund what a canceled job already consumed. The pattern FlowRunner flows follow is map first, spend deliberately, and put a person in front of the jobs big enough to matter.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Structured extraction into a working dataset
A revenue team maintains a list of target URLs: partner directories, pricing pages, product listings. The agent runs Extract and Wait with a JSON Schema defining exactly the fields that matter, then appends each extracted record to a tracking sheet in [Google Sheets](/integrations/google-sheets). Because the extraction is schema-driven rather than screen-scraped, layout changes on the source sites degrade gracefully instead of silently corrupting columns, and the agent flags any URL whose extraction came back incomplete.
Documentation site into the knowledge base
When a vendor's documentation matters to the team, the agent runs Crawl and Wait scoped with include-path filters to just the docs section, receives every page as clean Markdown, and publishes each into [Notion](/integrations/notion) as a structured knowledge base. Get Crawl Errors closes the loop: the agent reports which URLs were blocked by robots.txt or failed, so the team knows what the knowledge base does not contain, which is half of trusting it.
Research sweep with a spend gate
An analyst asks for the state of play on a topic. The agent runs Search Web with scraping enabled, pulls the full content of the top results, and posts a synthesized summary to [Slack](/integrations/slack) with source links. When the findings justify going deeper on a specific site, the agent maps it first, and if the crawl would be large, it brings the numbers to a person before starting the job rather than after finishing it.
Human-in-Loop Highlight
Start Crawl is where this connector can quietly spend a month's budget in an afternoon. Credits are consumed per page, a crawl follows links until it hits its limit, and Cancel Crawl explicitly does not refund credits already burned. So before any large crawl, the FlowRunner agent runs Map Website and Get Credit Usage, then stops and shows its work: "Mapping docs.vendor.com found 11,940 URLs. Current balance is 8,200 credits this billing period. Crawl the full site, limit it to the /api and /guides sections at roughly 2,100 pages, or skip?" The operator picks the scoped option, the agent sets the include-path filters and page limit accordingly, and the job runs inside a budget someone actually chose. This is [human-in-the-loop](/concepts/human-in-the-loop) applied to spend rather than safety: the agent brings the map and the meter, and a person decides how much web to buy.
Agent Capabilities
17 actionsScraping
4- Scrape URL Scrape a single URL synchronously into Markdown, HTML, links, screenshots, an LLM summary, or schema-extracted JSON, with JavaScript rendering, PDFs, proxies, and caching handled automatically. Supports browser actions such as click, scroll, and type before capture.
- Start Batch Scrape Run many URLs through the same scrape options as a tracked asynchronous job, with optional webhooks on completion.
- Get Batch Scrape Status Poll a batch job's progress counters and collect its chunked results.
- Batch Scrape and Wait The one-step synchronous version for small and medium URL lists: start, poll, and return all documents, with pagination followed automatically.
Crawling
6- Start Crawl Crawl a site from a start URL as an asynchronous job, guided by include and exclude path filters, page limits, sitemap handling, and crawl delay.
- Get Crawl Status Poll a crawl's progress and collect its results, which are chunked and expire about 24 hours after completion, so flows collect them promptly.
- Crawl and Wait Synchronous crawling for small jobs with a low page limit: start, poll until done, return every page.
- Cancel Crawl Stop a running crawl. Pages already scraped remain readable, and credits already consumed are not refunded, which is exactly why big crawls get approved before they start.
- Get Crawl Errors Diagnose why a crawl returned fewer pages than expected, including robots.txt blocks.
- Get Active Crawls Monitor every crawl currently running for the team.
Discovery and Search
2- Map Website Discover a site's URLs extremely fast from sitemaps and link discovery, optionally ranked by relevance to a search term. The reconnaissance step that makes every crawl a scoped decision instead of a guess.
- Search Web Search the web and optionally scrape the full content of every result in one call, with time, category, and geo filters. Titles and URLs only when speed and cost matter; full content when the analysis needs it.
Structured Extraction
3- Start Extract Run LLM extraction against one or many URLs as a tracked job, pulling structured data defined by a prompt or JSON Schema, with glob patterns covering whole site sections and web search enrichment filling gaps.
- Get Extract Status Poll an extraction job until the structured result is ready.
- Extract and Wait The synchronous version for a handful of URLs: extract, poll, and return the structured object in one step.
Usage
2- Get Credit Usage Read the remaining and total scraping credits for the billing period. The meter every large job gets checked against before it runs.
- Get Token Usage Read the extraction tokens consumed and remaining for the billing period.
Frequently Asked Questions
What can FlowRunner do with Firecrawl?
FlowRunner agents can run Scrape URL, Start Batch Scrape, and Get Batch Scrape Status in Firecrawl, plus 14 more actions.
Does connecting Firecrawl to FlowRunner require OAuth?
No. Firecrawl connects to FlowRunner with an API key, no OAuth flow required.
Can Firecrawl trigger a FlowRunner workflow automatically?
Firecrawl doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Firecrawl
$100 in credits. No card required. Connect in minutes.