What it is
Firecrawl MCP connects agents to Firecrawl, a web data API. Its core tools are firecrawl_scrape for one URL, firecrawl_map to list a site’s URLs, firecrawl_crawl for many pages, firecrawl_search for web search with optional page content, firecrawl_parse for PDFs and office files, and firecrawl_interact to click or type before reading. Newer additions cover change monitoring, research-paper, developer-docs and government search, and Alexandria, Firecrawl’s catalogue of paid data providers.
The hosted endpoint at https://mcp.firecrawl.dev/v2/mcp works with no key at all (scrape, search and parse only, rate-limited), with OAuth, or with an API key sent as a bearer header. It also runs locally with npx -y firecrawl-mcp or Docker, and can point at a self-hosted Firecrawl API. The npm package is on 3.34.1 (22 October 2026); GitHub releases stopped at v3.2.1 in 2025, so the commit log is the real changelog.
Install
https://mcp.firecrawl.dev/v2/mcpOur review
Firecrawl is the strongest general web-data server in this directory because it handles the jobs a plain fetch fails at, from JavaScript-rendered pages and whole-site crawls to schema-driven extraction and search with page content, under one credential. Its README is unusually good at telling the model which tool to pick, which matters more than it sounds when an agent has 27 to choose from.
That number is the main engineering cost. The full profile registers 27 tools, a heavy load in clients without tool search. Firecrawl offers two lighter doors: the keyless hosted endpoint exposes only scrape, search and parse, and a search-only endpoint exposes a fixed set of nine. Two of the 27 are feedback tools that exist mostly to refund a credit per search; environment flags remove them.
Credits are cheap per page and easy to burn. A basic scrape, crawl or map costs 1 credit a page, search costs 2 per 10 results, and JSON extraction adds 4 a page, so a 2,000-page crawl with structured output uses 10,000 credits, twice the Hobby allowance. Pages that return 403 or 404 still cost a credit. The agent tool defaults to a 2,500-credit ceiling per request unless you pass a lower maxCredits, which is the setting to change first.
Everything the server returns is untrusted web content headed straight into the model’s context, and the README describes no filtering of instructions hidden in pages. Zero data retention, SSO and API-key IP restrictions are Enterprise-only, while SOC 2 Type II covers every plan. Self-hosting keeps scraped data in-house; the README’s one sharp warning is to publish the Docker port on 127.0.0.1 only, because the container will spend its own API key for anyone who connects.
Where it earns its place
A two-person growth team at a B2B startup
Enrich a 5,000-row lead list with each company’s pricing and team size from its website
The agent scrapes each company’s pricing and about pages with JSON format against a schema and writes rows back. At two pages and 5 credits a page that is about 50,000 credits, which fits Standard at $83 a month billed annually; Hobby’s 5,000 credits would run out after 500 companies.
A developer running a support bot for a SaaS product
Keep the bot’s knowledge in step with a 1,500-page documentation site
One crawl costs about 1,500 credits; after that, firecrawl_monitor watches 50 key pages daily at 1 credit a page per check, roughly 1,500 credits a month. Together that fits inside Hobby’s 5,000 and avoids re-crawling the whole site for every release.
A research engineer on an AI evaluation team
Collect recent papers and the code repositories behind them for a literature review
The firecrawl_research_* tools search paper abstracts and full text and inspect GitHub repositories. Firecrawl says its research index endpoints are free in every category, so credits go only on the handful of pages scraped in full.
Before you commit
- Setup
- Add https://mcp.firecrawl.dev/v2/mcp as a remote server and sign in with OAuth, or send an API key as a bearer header; locally, run npx -y firecrawl-mcp (Node 22 or later) with FIRECRAWL_API_KEY, or the ghcr.io Docker image. The README gives configs for Cursor, Windsurf, VS Code and Claude Desktop.
- Integrations
- 27 tools in the full profile: scrape, map, search, crawl and status, parse, agent and status, interact, monitors, research, developer and government search, credit usage, feedback and Alexandria lookups. Writes happen only on your Firecrawl account (crawl jobs, monitors, agent runs), and all of them spend credits.
- Data and privacy
- The server forwards your key or OAuth token to Firecrawl’s API, which sees every URL and page it fetches; zero data retention is Enterprise-only, and a data processing agreement is countersigned from Standard upwards. Treat any instructions found in scraped pages as data, not commands.
- Cost at scale
- Billed annually, Hobby is $16 a month for 5,000 credits and Scale $599 for 1M; pay-as-you-go top-ups cost $5 for 1,000 to 5,000 credits depending on plan. Rate limits bite before credits on small plans: Free allows 10 scrape or search requests a minute and 2 crawls, Hobby 100 and 20.
- Limits
- Credits roll over only on Scale and Enterprise. The hosted parse tool cannot read your disk and uses a two-step upload, and local file parsing needs a self-hosted API. The agent tool is in preview, with 5 free daily runs and dynamic pricing.
What's good
- Scrape, crawl, map, search, parse and interact under one key
- Keyless hosted tier with 3 tools for quick trials
- Search-only endpoint and flags shrink the tool list
- SOC 2 Type II on every plan; self-hostable API
What's not
- 27 tools in the full profile crowd context
- JSON extraction costs 5 credits a page; error pages still bill
- Zero data retention and SSO only on Enterprise
- Untrusted page content goes straight to the model
Best for
Agents and developers who need dependable web content, from single pages to whole sites, and will tune tools and credit caps.
Alternatives
Firecrawl MCP: common questions
- Is Firecrawl MCP worth it?
- Firecrawl covers more of the web-data job than any rival server, but at 27 tools and per-page credits it rewards teams that configure it rather than just add it. We rate it 4.4 out of 5.
- How much does Firecrawl MCP cost?
- Free plan with 1,000 credits a month; Hobby $16/month billed annually ($19 monthly) for 5,000; Standard $83 ($99) for 100,000; Growth $333 ($399) for 500,000; Scale $599 ($749) for 1M.
- What are the downsides of Firecrawl MCP?
- 27 tools in the full profile crowd context; jSON extraction costs 5 credits a page; error pages still bill; zero data retention and SSO only on Enterprise; untrusted page content goes straight to the model.
- Who is Firecrawl MCP best for?
- Agents and developers who need dependable web content, from single pages to whole sites, and will tune tools and credit caps.
- How much does Firecrawl MCP cost for a team or in production?
- Billed annually, Hobby is $16 a month for 5,000 credits and Scale $599 for 1M; pay-as-you-go top-ups cost $5 for 1,000 to 5,000 credits depending on plan. Rate limits bite before credits on small plans: Free allows 10 scrape or search requests a minute and 2 crawls, Hobby 100 and 20.
- Is Firecrawl MCP safe for company data?
- The server forwards your key or OAuth token to Firecrawl’s API, which sees every URL and page it fetches; zero data retention is Enterprise-only, and a data processing agreement is countersigned from Standard upwards. Treat any instructions found in scraped pages as data, not commands.
- What are the best Firecrawl MCP alternatives?
- The closest alternatives we have reviewed are Apify MCP Server, Fetch (reference server) and Browserbase MCP.
Answers checked Oct 22, 2026.
Make Firecrawl MCP? Show this rating on your site
