Web Scraping MCP Servers Compared: Firecrawl, Apify & More

An MCP server lets an AI client (Claude Code, Cursor, VS Code and others) call a scraping tool directly instead of you pasting page text into a chat. Five scraping vendors ship one that matters. They differ in what the agent can do, how you authenticate, and what a run costs, so this guide compares them from each vendor's own docs, fetched on 2026-11-04.

Quick comparison

Firecrawl MCP

Firecrawl is the only one here that works with no account. Its docs list three ways in: keyless, browser sign-in, or an API key.

Keyless (Streamable HTTP, no credential):

claude mcp add --transport http firecrawl https://mcp.firecrawl.dev/v2/mcp

Keyless is rate limited and exposes firecrawl_search, firecrawl_scrape and firecrawl_parse. For the full tool set, add an API key as a bearer token on the same endpoint:

{
  "mcpServers": {
    "firecrawl": {
      "type": "http",
      "url": "https://mcp.firecrawl.dev/v2/mcp",
      "headers": { "Authorization": "Bearer <FIRECRAWL_API_KEY>" }
    }
  }
}

With an authenticated connection the docs also list firecrawl_map (discover URLs), firecrawl_crawl, firecrawl_agent (multi-source research), firecrawl_interact (clicks and form fills) and firecrawl_monitor_* (change tracking), subject to plan and team policy. To run it yourself, the local server needs Node.js 22 or newer and FIRECRAWL_API_KEY; set FIRECRAWL_API_URL to point it at a self-hosted Firecrawl API:

env FIRECRAWL_API_KEY=fc-YOUR-API-KEY npx -y firecrawl-mcp

Cost: calls use your Firecrawl credits. See the Firecrawl pricing table for plans.

Try Firecrawl ↗

Apify MCP

Apify's server is a gateway to the Apify Store: the agent searches for an Actor, reads its input schema and runs it. That makes it the broadest option for site-specific data such as Instagram or Google search results.

Hosted, with OAuth sign-in on first connection:

{ "mcpServers": { "apify": { "url": "https://mcp.apify.com" } } }

Local stdio, with an API token:

{
  "mcpServers": {
    "actors-mcp-server": {
      "command": "npx",
      "args": ["-y", "@apify/actors-mcp-server"],
      "env": { "APIFY_TOKEN": "YOUR_APIFY_TOKEN" }
    }
  }
}

By default it loads the actors and docs tool categories plus the apify/rag-web-browser and apify/web-fetch Actors. You can narrow it with a URL parameter, for example https://mcp.apify.com?tools=apify/instagram-scraper,apify/google-search-scraper. The core tools are search-actors, fetch-actor-details and call-actor; call-actor waits up to 30 seconds by default and returns the run status and storage IDs, not the data itself, so the agent then reads the dataset. Full-permission Actors are deliberately excluded.

Cost: you pay for the Actor runs you trigger, and each Actor sets its own price. Cap spend with maxTotalChargeUsd in callOptions. Details in our Apify review.

Try Apify ↗

Bright Data MCP

Bright Data's README says its server exposes 32 tools: search (Bing and Google), page scraping as Markdown with bot detection, CAPTCHA solving and proxy rotation handled on each request, structured data from 1,200+ pre-built scrapers, and a remote browser.

claude mcp add --transport http brightdata "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"

To run it locally, use npx @brightdata/mcp with API_TOKEN set. Trim the tool list with groups= or tools= URL parameters so the model isn't offered 32 tools at once, for example &tools=search_engine,scrape_as_markdown.

Cost: the README states 5,000 free requests per month, then pay as you go at $1.50 per 1,000 search/scrape/extract results and $8 per GB for browser navigation. Check Bright Data's MCP pricing page before relying on those numbers. See also our Bright Data review.

Try Bright Data ↗

Browserbase MCP

Browserbase gives the agent a real cloud browser, built on Stagehand. Instead of fetching a page, the agent calls tools like navigate, act ("click the login button") and extract, and can start and reuse sessions. Use it when the data sits behind interaction.

claude mcp add --transport http browserbase "https://mcp.browserbase.com/mcp?browserbaseApiKey=YOUR_BROWSERBASE_API_KEY"

The hosted endpoint is recommended in the docs; a local STDIO mode also exists. The API key goes in the URL, so treat that config like a secret and keep it out of version control. Cost is Browserbase's browser usage; see the Browserbase page.

Try Browserbase ↗

Crawl4AI MCP

Crawl4AI is open source, so its MCP server comes from the Docker server you run yourself. It exposes SSE at http://localhost:11235/mcp/sse and WebSocket at ws://localhost:11235/mcp/ws:

claude mcp add --transport sse c4ai-sse http://localhost:11235/mcp/sse

The documented tools are md, html, screenshot, pdf, execute_js, crawl and ask. The docs say versions 0.9.0 and later are secure by default, so read the self-hosting guide before exposing the port beyond localhost. There is no per-call fee, but you run and maintain the infrastructure yourself. Crawl4AI's site also advertises a hosted Crawl4AI Cloud with a free monthly credit allowance; we haven't reviewed it. More in our Crawl4AI page.

Which one should your agent use?

• Just need page text or search, and want zero setup: Firecrawl keyless, then add a key when you hit the limit.
• Need data from a specific site that already has a scraper: Apify.
• Pages block ordinary fetchers: Bright Data.
• Need clicks, logins or forms: Browserbase.
• No vendor, no per-call cost: Crawl4AI, if you'll host it.

Two practical tips regardless of vendor. Enable only the tools you need, because every tool description consumes context in each request. And treat scraped pages as untrusted input: a page can contain text written to manipulate your agent, so don't give a scraping-enabled agent tools that can send data out or spend money without confirmation.

Want typed JSON rather than raw page text? Compare the options in our best AI web scrapers ranking, or try the extraction playground.