Firecrawl MCP Server is the official Model Context Protocol server for bringing Firecrawl web data tools into AI coding agents and desktop clients. The current README positions it around live-web search, scraping, crawling, mapping, extraction, browser interaction, and agent-style research, rather than as a generic browser automation server. It can be used through the hosted remote endpoint at mcp.firecrawl.dev, through an API-key endpoint for higher limits and the full tool set, or by running the npm package locally inside Claude Desktop, Cursor, Windsurf, VS Code-style MCP clients, and other compatible hosts.
The most important 2026 change is the hosted MCP path. Firecrawl now documents a keyless free tier where scrape, search, and interact work without an API key but are rate-limited; crawl, map, agent, extract, and higher-limit usage still require an API key or OAuth. That makes the server easier to trial, while production teams still need to model Firecrawl API credits, concurrency limits, retries, and credit monitoring. The upstream repo also documents cloud and self-hosted support, including a custom FIRECRAWL_API_URL for teams that run Firecrawl inside their own environment.
Pricing should be read as Firecrawl platform pricing, not as a separate MCP-server subscription. The public pricing page lists a Free plan with 1,000 credits per month, Hobby at $16/month billed yearly, Standard at $83/month billed yearly, Growth at $333/month billed yearly, Scale at $599/month billed yearly for 1,000,000 credits, and Enterprise custom pricing. The GitHub repository is MIT-licensed, active, and had about 6.7K stars at verification time, but reliability, anti-bot handling, and credit burn still need workload-specific testing before a review page should make hands-on claims.
