Most servers here stop at fetching a page. This one carries search, crawl, a live browser session, scheduled change monitoring and a research index on the same connection, so a task that starts as "read this page" can become "watch it and tell me when it changes" without another install.
Firecrawl's own MCP server. It turns web pages into clean, agent-ready content — one URL scraped, a whole site crawled, a search across web, news or images — and goes past retrieval into live browser interaction, recurring change monitors, and a research index over papers and GitHub. Available as a hosted endpoint or run locally with npx.
- One URL retrieved and extracted, or a document parsed into markdown, HTML, links, a summary, or JSON matching a schema you supply (`firecrawl_scrape`, `firecrawl_parse`)
- Web, news and image search with operators — quoted phrases, `-term`, site filters — plus feedback tools for telling it when a result was wrong (`firecrawl_search`, `firecrawl_search_feedback`, `firecrawl_feedback`)
- A site's indexed URLs enumerated without fetching each page (`firecrawl_map`)
- A multi-page crawl started, polled to a terminal state, and its collected results read (`firecrawl_crawl`, `firecrawl_check_crawl_status`)
- A live browser session to navigate, click controls, fill fields or run code on a page, then released (`firecrawl_interact`, `firecrawl_interact_stop`)
- An asynchronous research job from a prompt, optional seed URLs and an optional JSON schema (`firecrawl_agent`, `firecrawl_agent_status`)
- Recurring monitors that compare each check with the previous one, reporting per-page same or changed, with a history of checks (`firecrawl_monitor_create`, `firecrawl_monitor_run`, `firecrawl_monitor_checks`, `firecrawl_monitor_list`, `firecrawl_monitor_update`, `firecrawl_monitor_delete`)
- A paper corpus searched in natural language, papers resolved from an arXiv, PMC, PMID or DOI identifier, citation-graph neighbours from seed IDs, and in-body passages read against a question (`firecrawl_research_search_papers`, `firecrawl_research_inspect_paper`, `firecrawl_research_related_papers`, `firecrawl_research_read_paper`)
- Indexed GitHub issue, pull-request and README content searched, returning ranked matches (`firecrawl_research_search_github`, `firecrawl_developer_search`)
Nothing at all for a first look: the hosted endpoint at `https://mcp.firecrawl.dev/v2/mcp` runs scrape, search and parse on a rate-limited keyless tier. Crawl, map and agent need a Firecrawl API key. Give the key to the client's secret or header setting as `Authorization: Bearer <key>` — never in the server URL and never in a chat. For an interactive sign-in there is an OAuth endpoint at `https://mcp.firecrawl.dev/v2/mcp-oauth`, and a read-only surface at `https://mcp.firecrawl.dev/v2/mcp-search` that exposes only search and the five research tools and fetches no page content. Running it locally is `npx -y firecrawl-mcp` with `FIRECRAWL_API_KEY` set, or `FIRECRAWL_API_URL` pointed at a self-hosted instance.
One command plus a key — env FIRECRAWL_API_KEY=fc-YOUR_API_KEY npx -y firecrawl-mcp, then supply credentials