Firecrawl

Firecrawl turns a page or a whole site into clean markdown instead of raw HTML. That matters for anything that reads the result afterward, since markdown strips the layout noise a model would otherwise parse around. JavaScript is executed first, so a page that builds itself in the browser comes back as the page a person sees.

Does Firecrawl have an MCP?

Yes. Firecrawl hosts its own MCP server and documents three ways in: a keyless trial limited to Search, Scrape and Parse within daily limits, a browser sign-in at https://mcp.firecrawl.dev/v2/mcp-oauth, and an API key sent as a bearer token to https://mcp.firecrawl.dev/v2/mcp. It also offers native connectors for Claude, ChatGPT and Codex. Tulina reaches Firecrawl through its own MCP server instead, so the scraped markdown lands in the same conversation as the search that found the page and the CRM that will hold the result.

How do you connect Firecrawl to Claude and ChatGPT?

1. In Firecrawl, create an API key on the API keys page. It starts with fc-.
2. Paste it into Tulina's Firecrawl connector. It is stored in the vault and never shown to the model. There is no shared Tulina key, so the credits are yours.
3. Add https://mcp.tulina.ai/mcp to Claude or ChatGPT and sign in with your Tulina account. The MCP page has the same link.

Several Firecrawl accounts can be stored as named accounts, and a request names the one it is for.

Tulina vs Firecrawl's official MCP server

Both put Firecrawl's scraping in front of a model. The differences are in how you sign in and what else shares the connection.

If Firecrawl is the only tool the model needs, or you want the keyless trial, Firecrawl's server is the shorter path. Tulina earns its place when a scrape is one step in a longer piece of work.

What you can ask it

"Map docs.example.com, then pull the pricing and changelog pages as markdown." The agent runs a map first, which returns URLs only and costs little, picks the two pages from the list, and scrapes each one. Mapping before reading keeps the credits on the pages that matter.

"Crawl the blog on example.com, 50 pages at most, and summarise the last six posts." A crawl is asynchronous, so the agent starts it with a page limit, then checks the job until it reports completed and reads the pages it returned. The limit matters: the API default is 10,000 pages, and every rendered page uses credits.

"Pull plan name, monthly price and seat limit from these twelve pricing pages into a table." The agent starts an extraction with a schema covering the three fields, checks the job until it finishes, and returns one row per page. For a single page it uses a scrape with a JSON schema instead, which answers in one call.

Processes chain these steps without a prompt each time. Message new LinkedIn connections reads each person's site before it writes, and the full list is below. Firecrawl pairs naturally with Serper for finding pages and Apify for platforms that need a dedicated scraper.

What the Firecrawl connector can't do

• It exposes eight tools: scrape, map, crawl, crawl status, cancel crawl, search, extract and extract status. Nothing outside them is reachable through Tulina.
• It needs your own Firecrawl API key. Tulina holds no shared key, and each account pays Firecrawl for its own credits.
• A page behind a login is out of reach. Tulina's browser connector covers that case.
• A site that blocks the default fetch needs the stealth proxy option, which costs more credits, so it is not set by default.
• Crawls and extractions run as jobs. The first call returns an id and the results arrive on later status calls.
• Under a Tulina project that fences off URL prefixes, scrape, map, crawl and extract refuse a matching address, and search, map and crawl-status results from those prefixes are dropped.
• Raw HTML without JavaScript rendering is a job for Serper's scrape tool, which is enough on its own.