Firecrawl Review

Firecrawl is a web data API for AI systems. It searches the web, scrapes pages, crawls whole sites, and returns clean Markdown, HTML, screenshots, or JSON that follows a schema you define. It handles JavaScript-rendered pages, parses PDFs and DOCX files, and ships SDKs for Python, Node.js, Go, Rust, Java, and Elixir, plus a CLI and an MCP server for clients such as Cursor, Claude, and Windsurf.

Self-hosting uses Docker Compose from the repository and needs Git, Docker with Compose v2, and a free port 3002. The stack runs the API, PostgreSQL, Redis, RabbitMQ, workers, and a Playwright service, and it comes up on http://localhost:3002. The official guide says the baseline has no durable storage volumes, TLS, high availability, or authentication, so it is meant for trusted networks until you harden it. The guide does not publish a minimum host size.

Firecrawl is an API and SDK product rather than an app, so the experience is a request and a response: send a URL or a search query and receive Markdown or JSON. The hosted dashboard manages API keys and usage. We did not run the self-hosted stack ourselves; this section reflects the official documentation.

• Scrape any URL into Markdown, HTML, screenshots, metadata, or schema-based JSON
• Search that returns full-page Markdown with results, so no separate scrape step
• Interact mode for clicking, scrolling, typing, and multi-step flows such as logins
• Site crawling plus PDF and DOCX parsing
• MCP server, CLI, and SDKs for six languages

• By far the largest community in this space, with about 190,000 GitHub stars and daily commits
• Output designed for LLMs, which suits RAG pipelines, research agents, and data enrichment
• A generous free cloud tier and an easy path from hosted to self-hosted

• Self-hosted Firecrawl lacks Fire-engine, so screenshots, page actions, and advanced anti-bot handling are not in the default stack
• Agent, Browser, and some specialized formats are Cloud-only or need extra services
• AGPL-3.0 means a modified version offered over a network must share its source
• LLM-based extraction in a self-hosted setup needs an OpenAI-compatible provider or Ollama

Self-hosting is free apart from your infrastructure. Cloud plans are credit-based: Free gives 1,000 credits per month with no card, Hobby is $16/month billed yearly for 5,000 credits, Standard is $83/month for 100,000, Growth is $333/month for 500,000, and Scale is $599/month for 1,000,000. Enterprise is custom.

Jina AI and Firecrawl both turn web pages into clean text for LLMs, but they are different products at the edges. Jina is a broader search foundation with embeddings, rerankers, and a reader API, while Firecrawl focuses on scraping, crawling, search, and structured extraction, and lets you run the core yourself. If you only need to read single URLs, Jina's reader is lighter; if you need crawling and structured output, Firecrawl goes further.

• Teams building RAG pipelines, research agents, or lead-enrichment tools that need clean web data
• Developers who want a hosted API first and the option to self-host later
• Projects that need crawling and schema-based extraction, not just single-page reading

• Sites that need heavy anti-bot handling, unless you use Firecrawl Cloud
• Teams that need managed browser sessions for automation, where a browser infrastructure product fits better
• Organizations that cannot accept AGPL-3.0 obligations on modified deployments