This page is what happened when we ran it. The project itself — code, maintenance, community — has its own review: full repo review →

WebMCP servermeasured 19 Aug 2026

Firecrawl

Firecrawl's MCP server, published by the Firecrawl team as the npm package firecrawl-mcp, gives an AI agent web scraping, crawling, search, structured extraction, page monitoring and a small research toolkit over the hosted Firecrawl cloud. We started it and it registered 25 tools, reporting itself as firecrawl-fastmcp version 3.24.0 on protocol 2025-06-18.

Firecrawl · protocol 2025-06-18 · v3.24.0 · measured by Keyoor

BootBOOTS ✓We started it and initialize came back in 16.3s.

Tools25tools/list payload 36,312 bytes

Context tax~9,077estimated tokens for the whole tools/list, chars/4 estimate

Resources / prompts— / —counts returned by resources/list and prompts/list

Measured19 Aug 2026one run of the harness, clean environment, no credentials · how we test

How we started it

npx -y firecrawl-mcp

Verdict

Use it if scraping is central to what your agent does, because the tools work and the keyless mode lets you try it before paying anyone. The cost is real, though: 25 tools weighing roughly 9,077 estimated tokens sit in every prompt, and most sessions touch two of them. If your agent scrapes a page once a week, load a trimmed tool set or skip it and call the HTTP API directly.

Use it if

Your agent's main job is reading the live web and you want markdown back instead of raw HTML you have to clean up

You need crawl plus check-status as a pair, so the agent can start a multi-page job and come back to it later

You want to try scraping and search before creating an account, which the keyless mode allows at the price of per-IP rate limits

You are already paying for Firecrawl cloud and would rather not hand-write another HTTP client

You want page-change monitoring driven by the agent rather than by a cron job you maintain

Skip it if

Web access is a side errand in your agent, in which case 25 tool descriptions in context every turn buy you very little

Your client cannot filter or disable individual tools, because the full list is one of the heavier ones we measured

You need everything working without an account: the stderr notice says scrape and search are the free keyless pair and the rest wants FIRECRAWL_API_KEY

You cannot send target URLs to a third-party cloud, since the default setup does the fetching on Firecrawl's side, not yours

You care about cold start: it took 16,327 ms to initialize under npx on our box, which is slow enough to notice

You only need one fixed site scraped on a schedule, where a twenty-line script beats a tool surface this wide

The bill for your context window

Tool	Est. tokens	Described?	Required params
firecrawl_search	~1,411	yes (1177 chars)	1
firecrawl_crawl	~972	yes (476 chars)	1
firecrawl_scrape	~951	yes (1110 chars)	1
firecrawl_interact	~865	yes (456 chars)	0
firecrawl_parse	~674	yes (838 chars)	1
firecrawl_monitor_create	~401	yes (640 chars)	0
firecrawl_research_search_papers	~396	yes (466 chars)	1
firecrawl_developer_search	~289	yes (430 chars)	1

The tools/list payload measured 36,312 bytes, about 9,077 tokens at the chars/4 estimate, across 25 tools. The weight is lopsided: firecrawl_search alone accounts for roughly 1,411 estimated tokens because its description documents every source and filter, while firecrawl_monitor_get costs about 110. Search, crawl, scrape and interact are the expensive four; the six monitor tools and the five research tools are cheap individually but there are eleven of them. If your client supports tool filtering, keeping scrape, search, map and crawl gets you most of the value for a fraction of the bill.

Setup reality

Run it as npx -y firecrawl-mcp and it starts with no configuration at all. With no FIRECRAWL_API_KEY and no FIRECRAWL_API_URL, its own stderr tells you it is in keyless mode: firecrawl_scrape and firecrawl_search work free against the cloud with per-IP rate limits, and everything else wants a key. Set FIRECRAWL_API_KEY for crawl, extract, monitors and the agent tools; FIRECRAWL_API_URL points at a self-hosted instance instead. Transport is stdio. Our clean-env boot printed a deprecation warning from a transitive koa-router dependency and started anyway. The 16,327 ms initialization is mostly npx fetching the package, so a pinned local install will behave better.

Questions people ask

Does the Firecrawl MCP server need an API key?

Not to boot. We started it with a clean environment and it came up with all 25 tools listed. Its stderr says scrape and search run keyless against the Firecrawl cloud under per-IP rate limits, and the remaining tools need FIRECRAWL_API_KEY.

How much context does Firecrawl MCP use?

Its tools/list came to 36,312 bytes, an estimated 9,077 tokens at chars/4, measured 2026-08-19. That is charged on every request your client sends, not once per session, so tool filtering pays for itself quickly here.

Can I point it at a self-hosted Firecrawl?

Yes. FIRECRAWL_API_URL redirects the server at your own instance, which is the route to take when the URLs you scrape cannot leave your network. The tool surface stays the same size either way.

Why is first boot so slow?

We measured 16,327 ms to initialize, and most of that is npx resolving and downloading firecrawl-mcp on first run. Install the package locally and point your client at the binary if a slow client startup bothers you.

Other web servers we started

Server	Vendor	Boots	Context tax
Apify Actors	Apify	✗	~0 tokens
Fetch	Anthropic (reference)	✓	~276 tokens

How this page is made: the server is spawned as an unprivileged user with a clean environment and no credentials, then asked for its tools, resources and prompts over stdio. Token figures are estimates at four characters per token, not a tokenizer count. One run, one machine. Corrections: contact the desk.

← All measured MCP servers