Firecrawl
Firecrawl is a web data API for AI. Its website says one API can search, scrape and interact with the web. It turns messy sites into data that language models can use, as Markdown, JSON or screenshots. The code is open source on GitHub, and the site cites more than 180,000 stars.
What Firecrawl offers
Features and credit rules come from the Firecrawl website, its pricing page and developer docs, read on October 24, 2026. Firecrawl's current prices took effect on September 4, 2026; if they change again, the pricing page wins.
| Firecrawl capability | What Firecrawl states | Where it fits |
|---|---|---|
| Search | Send a query and get relevant results, each with full-page Markdown already included. 2 credits per 10 results | Research agents and retrieval that starts from a question |
| Scrape, Crawl and Map | Turn URLs into clean Markdown or JSON, with JavaScript rendering. A basic scrape costs 1 credit per page | RAG knowledge bases and loading site content |
| Interact | Click, scroll and type to reach content that needs actions. 2 credits per browser minute | Pages that need multi-step actions |
| Pricing | Free tier of 1,000 credits a month with no card. Hobby is $19 a month for 5,000 credits, Standard $99 for 100,000, Growth $399, Scale $749 for 1,000,000. Prices apply from September 4, 2026 | Start free, then move up by credits |
| Docs and open-source repo | SDKs, a command-line tool, an MCP server and skills for tools such as Claude Code. The code is on GitHub | Self-host it or use the hosted version |
The pricing page also says a scrape that returns no result is not charged. A page that returns an error such as 403 or 404 still costs 1 credit.
Who it is for
- Developers building AI agents and RAG. They turn pages into Markdown that uses fewer tokens. The site claims about 93% fewer input tokens than raw pages.
- Teams doing competitor monitoring and lead enrichment. They scrape public pages on a schedule and extract fields.
- Teams that want to self-host. The core is open source. You can run it yourself or use the hosted version, which the site says has its own proxy and rendering infrastructure.
- Users who want direct links to AI tools. The site offers an MCP server that works with Cursor, Claude and other clients.
A support bot knowledge base: docs through Firecrawl, user questions through EveryInfra
A software product's support team or AI assistant needs two kinds of knowledge: what the product officially says (docs, help center, changelog) and what users actually ask (community threads and Q&A). The first lives on your own or a competitor's site; the second lives on discussion platforms.
- Official content. Run Firecrawl Crawl over the help center to turn the whole site into clean Markdown, then chunk it into a vector store. Pages rendered with JavaScript are fine, because Firecrawl renders JS.
- User questions. The EveryInfra Reddit post detail and comments and Zhihu questions and answers endpoints return real questions, answers and upvote counts by product name or keyword. The most upvoted questions are often the entries your knowledge base lacks.
- Find the gaps. Ask a model to compare the frequent community questions with the docs, one by one, and list what users ask that the docs do not cover. Hand that list to the docs team.
- Keep it fresh. Rerun weekly. On the docs side, diff old and new Markdown and update only the entries that changed; on the community side, take only threads posted since the last run.
On cost, Firecrawl does not charge credits for a scrape that returns nothing, and EveryInfra does not charge for failed or empty calls. Both have MCP servers that can run side by side in Claude Code or Cursor; EveryInfra's connection steps are in the MCP guide.
Crawl one help center on the Firecrawl free tier
The Firecrawl free tier gives 1,000 credits a month, and a basic scrape costs 1 credit per page, which is enough to crawl a mid-sized help center once. You can also self-host from GitHub. Credit costs per feature are listed at firecrawl.dev; check them there before a large crawl.