Firecrawl's Grok Bot plugin: what's free, and what the 21% claim means

Technology

Quick answer: Firecrawl’s Grok Bot plugin is real and official. Firecrawl’s changelog lists it on Nov 19, 2026. The x.ai plugin page says it was published by Firecrawl on the Cursor Marketplace, and the marketplace lists it as “Verified by Cursor” with 1 MCP connector and 19 skills. The free plan gives 1,000 credits every month, refreshed monthly, with no credit card. Unused free credits do not roll over. When the balance hits zero, requests that need credits return HTTP 402 until the next monthly reset. The viral “500 free searches” line only holds if every search stays at 10 results or fewer and you scrape nothing. A more typical search-plus-five-pages job costs about 7 credits, or roughly 142 of those jobs a month. Firecrawl’s “21% better answers” figure is Firecrawl’s own blind test across 845 tasks, with no published method. Connecting the plugin means Firecrawl sees every URL, query, and page it fetches for you. Zero Data Retention is a paid / Enterprise-style add-on, not the free default.

Who this is for: Anyone who runs a Grok Bot and saw a post about adding Firecrawl. You are deciding whether to connect a scrape/search app, not whether your PC was hacked.

Who this is not for: People who never use Grok Bot connectors. You can skip it.

What launched

On Nov 19, 2026, Firecrawl published “Introducing the Firecrawl plugin for Grok Bot” on its changelog. The listing on x.ai/bot/plugin/789 says: “Published by Firecrawl on the Cursor Marketplace.” The Cursor Marketplace page shows “Verified by Cursor”, MCP 1, and 19 skills. The marketplace copy says it “Runs on the hosted Firecrawl MCP server with browser sign-in, no API key to paste.”

What it can do, per Firecrawl and the marketplace listing: search the live web, scrape a page into clean text or structured data, crawl sites, map URLs, and run interactive browser sessions. Firecrawl also bundles Alexandria, its library of licensed data providers and indexes, behind the same connection.

Add it once under Connect apps and it is account-wide. x.ai docs say: “Apps are account-wide. A connected app is available to every Bot you run.” Routines that use a connector also depend on that sign-in. So every Bot and routine on the account shares one Firecrawl credit pool.

What “free” actually means

Firecrawl’s pricing page (effective wording checked Nov 20, 2026):

Firecrawl’s billing docs say free-plan requests that need credits return HTTP 402 (Payment Required) once the monthly allotment runs out, until the monthly reset. The pricing table frames Free as “500 searches or 1,000 pages scraped” from one shared balance, not two separate piles. Those headline numbers use the base rates only.

Base rates from Firecrawl’s billing docs and pricing page:

Action Credits
Basic scrape or crawl page 1 per page
Search 2 per 10 results (rounded up; 11 results = 4)
Search result that is also scraped search cost + scrape cost per page
Question / query format on a page +4 per page
JSON / structured extract +4 per page
Interact (browser session) 2 credits per browser minute code-only; 7 per minute when driven by a prompt (one-minute minimum)

So the “500 free searches” marketing math is: 1,000 ÷ 2 = 500, and only if each search returns 10 or fewer results and you never scrape. A launch-style job that searches and then reads five result pages costs about 2 + 5 = 7 credits, or about 142 such jobs per free month (1,000 ÷ 7). Crawl a 200-page docs section and you spend 200 credits, a fifth of the free month, before any extras.

Free-plan speed limits (pricing page): 2 concurrent browsers; 10 requests a minute for scrape, map, and search; 2 a minute for crawl, agent, and interact.

Paid plans start at Hobby ($19/month or $16/month billed yearly for 5,000 credits). Paid plans can enable pay-as-you-go in $5 increments. That is a separate spend decision. This article is about the free tier.

The 21% claim

Firecrawl’s changelog and marketing say the plugin can “lift answer quality 21%” through Alexandria. Firecrawl’s own Alexandria page states, word for word: “In our own blind test across 845 tasks, agents with Alexandria scored 21% higher on answer quality than the same agents with built-in web tools.”

What that is not:

Treat 21% and 845 tasks as Firecrawl’s own unpublished evaluation until a method appears.

Firecrawl also markets “3,300+ data tools” on the Nov 19 changelog. Firecrawl’s own counts have varied (for example “3,000+ tools” on an earlier changelog and “550+ data providers” in other posts). Tools and providers are not the same unit. Label the 3,300+ figure as Firecrawl’s count.

“My usage meter didn’t move” is the wrong test

Some early posts said Firecrawl credits are separate, so a Grok Bot usage meter “didn’t move” after a small job. Two meters are not one free lunch.

A tiny job can be too small to move a weekly percentage meter. That is not proof that heavy Firecrawl work is “free” on the Bot side. Clean text may use fewer tokens than clicking through pages in a browser, but that trade-off has not been measured publicly for Grok Bot.

Privacy: what Firecrawl sees

When a Bot calls Firecrawl, Firecrawl’s service receives the URLs, search queries, and questions you send, and the page content it fetches. Firecrawl’s privacy policy (last revised December 26, 2024) says its servers are in the United States, and that data “could pass through other countries as part of the transmission process.”

On retention, the same policy says personal information is kept “until you request in writing that we delete” it, and that “we do not currently have” a scheduled deletion policy. The policy does not spell out a separate automatic wipe for every scrape query.

Zero Data Retention (ZDR) is not the free default. Firecrawl’s pricing page lists Zero Data Retention as an Enterprise feature. Billing docs also price a ZDR scrape option at +1 credit per page, and note that some ZDR search settings are enabled per team alongside Enterprise-style features. If you need “fetch and forget,” check your plan before you send sensitive URLs.

Practical rule: do not send a connector any URL, login-gated page, or customer detail you would not hand to a third-party SaaS.

What to do

  1. Decide if you need it. Built-in web tools already cover many one-off lookups. Firecrawl helps most when you need structured scrapes, whole-section crawls, or Alexandria’s licensed sources.
  2. If you try the free plan: add Firecrawl from the Grok Bot marketplace, sign in (that creates or uses a Firecrawl account), and watch the Firecrawl credit balance. Expect HTTP 402 when credits hit zero. No card means it fails closed instead of charging you.
  3. Budget the shared pool. One connection serves every Bot and routine on the account. Agree who may burn credits before a crawl job eats the month.
  4. Do not use a quiet usage meter as proof it is free. Check Firecrawl’s credit ledger and your Bot usage separately.
  5. Treat the 21% line as marketing until Firecrawl publishes the method, sample, and scoring.
  6. Keep sensitive work off it unless you have ZDR (or an equivalent contract) and a clear need.

Aside for readers who run several bots: teams that already automate site audits, SEO briefs, or competitor research may find a shared scrape/search connector useful later. That is a product fit note, not an install checklist, and it is not a claim that any specific newsroom has turned it on.

What we still don’t know

Sources

AltTab Dad