Firecrawl Reviews & Product Details

I use Firecrawl as a tool inside an agent pipeline rather than as a manual scraper, and that's where it earns its place. The API surface is small /scrape, /search, /map, /extract, /crawl, so wiring it into a workflow takes minutes and I don't have to write per-site parsers or babysit a headless browser.

What stands out: the markdown output is genuinely clean, with headings, tables and code blocks intact, which matters when the text gets fed into a model rather than read by a human. /map gives me a usable URL inventory of an unfamiliar site in one call, and search-plus-scrape removes a whole round of plumbing I used to build myself. Failure modes are legible too, when a page returns nothing I can usually tell whether it was JS rendering, a block, or a genuinely empty page, which is rarer in this category than it should be. Review collected by and hosted on G2.com.

Credit accounting is the weakest part for me. It isn't obvious up front what a crawl of N pages will actually consume, so I stay conservative with page limits and check the dashboard rather than trusting an estimate, that caution costs me coverage on large sites.

Second, per-page control is thinner than I'd like: fewer knobs for pagination, per-page timeouts and retry behaviour on slow or flaky domains. On a site that times out intermittently I end up re-running the whole crawl instead of the pages that failed. Neither is a blocker, but both are the friction points I hit most often. Review collected by and hosted on G2.com.

I use Firecrawl through its MCP connector inside Claude, and the best part is that it returns structured data rather than raw pages. One call can run up to ten searches against a data provider and give back clean records with titles, dates, links and full descriptions, so I don't have to write or maintain my own scrapers. Large results are kept on their side, and I can filter them with jq before pulling anything into my session, which keeps my context small. Finding the right tool is free, and each call shows its credit cost up front, so I always know what a search will cost me before I run it. Review collected by and hosted on G2.com.

The credits go quickly on heavier research. A ten-search call costs 50 credits, and a few rounds of that plus the follow-up detail lookups can use up the daily allowance in one sitting. Stored results expire after five minutes of inactivity, so I often have to reload them mid-task. The built-in jq also stops at 10,000 iterations, which means slicing large results into chunks by hand. Results in a batch don't always come back in the order I requested them, so I've learned to check which result belongs to which query. Review collected by and hosted on G2.com.

Firecrawl fits into the tools we already run without any custom work. It is the page-reading backend in Hermes, with SearXNG handling search, and Open WebUI uses it in the same way. Searching, reading and answering happen in one chain, so I rarely have to move data between tools by hand.

What I value most is how cleanly it pulls the data. In October I captured a full industry report, about 19,300 words, from a web page into our engineering folder. The headings, figures and reference links came through intact, and the output is structured Markdown rather than page clutter. That matters for us because the AI is only as good as the content it reads. Getting clean, structured text back means I can trust the answers I get from our own queries, and I can check them against the source when it matters.

The web interface is clear, and the credits are easy to see at a glance, so there is never any doubt about how much we have left. For a startup SME the allowance is very generous, which leaves plenty of room to keep researching without rationing every query. Review collected by and hosted on G2.com.

As for what I dislike, there is very little. The one gap I have hit is difficulty scraping Reddit. That limits community research, such as following discussions from people who use the same kit or sector. It is a narrow gap and does not outweigh the rest of the experience. Review collected by and hosted on G2.com.

I like how Firecrawl simplifies web scraping and data extraction by converting websites into clean, structured, LLM-ready data. Its ability to crawl multiple pages, handle JavaScript-heavy websites, and integrate with AI workflows through APIs and MCP makes it particularly useful for research, automation, and building AI agents. It saves considerable time compared to building and maintaining custom scrapers. Review collected by and hosted on G2.com.

The main drawbacks are credit consumption and inconsistent results on complex or heavily protected websites. Large-scale crawling can become expensive, and some pages require additional configuration to extract information reliably. Better transparency around credit usage, more consistent handling of dynamic websites, and stronger debugging tools would make the experience significantly better. Review collected by and hosted on G2.com.

I like how Firecrawl brings website and job research into the same conversation with my AI assistant. The new Alexandria job-search tools are a promising addition. We tested a search and got structured listings with useful details and direct links. If you use AI for research or build agents, I’d recommend giving it a try. Review collected by and hosted on G2.com.

I need to keep an eye on credit usage when running several searches, and I still check job listings with the original employer before acting on them. Clearer visibility into how much a search will cost would make it easier to plan usage. Review collected by and hosted on G2.com.

It's one tool for all the web stuff my AI agents do. Web search, reading pages (even the JS-heavy ones), pulling what people say on Reddit, YouTube and X, fact-checking, and checking a brand's site before I make a sponsored video. It gives back clean markdown my AI can actually read, so I don't need a few separate tools for this. I use it every day, and our usage went from about 1,100 credits in August to 7,650 in October. Review collected by and hosted on G2.com.

The credit math takes a minute to figure out. Pages that come back with an error like a 403 or 404 still cost a credit, which adds up when an agent hits a lot of blocked sites. And there's a big jump between Hobby and Standard. We outgrew Hobby in two months, but Standard is way more than we need right now. Review collected by and hosted on G2.com.

Their search and scrape APIs help us build context around our customers, and we use that context to power various agents we deploy for them. Their MCP integration is also helpful as my coding agent can directly talk to it. Review collected by and hosted on G2.com.

The credit consumption per activity is high, and I would appreciate it if they could optimize for it further so that the overall costs are low as well as optimize on how coding agents consume tokens. Review collected by and hosted on G2.com.

What I like best is how much cleanup it saves us. We run an online academy and research lesson topics often. Firecrawl searches the web and turns pages into clean, readable text, without menus and ads, so we can go straight to the content. It also works smoothly inside Claude, so the research happens in the same place where we write our lessons. Review collected by and hosted on G2.com.

The daily credits ran out in the middle of a research session. There was no warning before they ran low, so we had to stop and switch to another tool. A clear alert when credits are getting low, or an easy way to see how many are left before a big search, would make it much easier to plan our work Review collected by and hosted on G2.com.

It turns any page into clean, LLM-ready markdown with almost no setup. I mostly use it through the MCP server inside Claude Code, where my agents can search, scrape and map a site in one workflow without me writing custom scrapers. Search with built-in highlights saves a lot of tokens, and the structured data providers make it easy to pull the same fields across many pages. Review collected by and hosted on G2.com.

Credit usage can be hard to predict, especially for crawls or for searches that scrape the full page of every result, so I have to watch my settings closely. Some JavaScript-heavy pages still need extra wait or action settings before they come back complete. The MCP tool list has also grown large enough that agents sometimes pick the wrong tool unless you tell them which one to use. Review collected by and hosted on G2.com.

It reliably gets pages that ordinary scrapers can’t. We run a nightly B2B lead-generation pipeline, and Firecrawl is our fallback when plain requests/BeautifulSoup hits JavaScript-rendered sites, WAF pages or 403 errors. Getting clean markdown, html and rawHtml back from one 1-credit call means we can use the markdown for classification and still pull addresses and phone numbers from the raw footer HTML. The API is simple and the credit-usage endpoint makes it easy to keep an eye on spending. When v1 was retired, moving to v2 only took changing the endpoint URL, because the response format stayed the same. Review collected by and hosted on G2.com.

At first the pricing for the different output formats wasn’t obvious to us. JSON/extract costs several credits per page, compared with 1 credit for a plain scrape. We didn’t notice that until our usage went over the plan and auto top-up kicked in. Our own log-based estimate was far below the real usage shown in the dashboard. A clearer per-request cost breakdown in the response or the dashboard, or an optional spending alert before auto top-up charges, would help. Some sites with strong bot protection still fail occasionally, but that’s expected for any scraping tool.

Both answers are based on what actually happened in our pipeline: the fallback scraping, the three formats in one call, the extra cost of extract and the auto top-up. Feel free to shorten them. Review collected by and hosted on G2.com.