Firecrawl

Firecrawl is an open-source and hosted API for search, scrape, crawl, map, and browser-style interact. You send a URL or query; it handles proxies, JS rendering, and extraction, then returns markdown, structured JSON, screenshots, or media-derived text. Teams use it so agents spend tokens on reasoning instead of cleaning HTML.

Most agent stacks still stall on messy HTML, JS-heavy sites, and one-off scrapers. Firecrawl is becoming the default “get me usable web context” layer — search, scrape, crawl, and interact behind one contract. If you ship product agents that need live web data, this is the project to track.

Call scrape or search via REST/SDK with an API key (or self-host). For multi-page jobs use crawl/map; for dynamic flows use interact after a scrape id. Agents can also attach via MCP (`firecrawl-mcp`) or a one-command skill install so tools appear inside the coding agent.

Not just another HTML downloader: reliability on JS sites, LLM-oriented output formats, and first-class agent/MCP paths in the same product. Hosted speed and coverage matter when a naive Playwright script does not scale.

• Public demo available
• API / SDK surface
• MCP server / client support

• Page Agent
• Browserbase ↗
• Jina Reader ↗