Web Scraping MCP Servers Compared (2026): Bright Data, Firecrawl, Apify, Tavily and Browserbase

• Bright Data: anti-bot, CAPTCHA solving and proxy rotation stated in its MCP docs; search, Markdown scraping, 13 browser tools and structured web_data_* extractors. Free: 5,000 requests/month, no card. Paid: pay as you go from $1.50 per 1,000 requests.
• Firecrawl: scrape to Markdown or schema JSON, crawl, map, search, a research agent, and firecrawl_interact for browser actions. Free: 1,000 credits/month. Paid: Hobby, $19/month for 5,000 credits.
• Apify: the agent searches the Apify Store and runs ready-made scrapers (Actors) with search-actors and call-actor, plus a RAG web browser. Free: $5 of usage/month. Paid: Starter, $19/month.
• Tavily: web search and page extraction built for agents (tavily-search, tavily-extract), plus map and crawl. Free: 1,000 credits/month. Paid: $30 for 4,000 credits, or $0.008 per credit pay as you go.
• Browserbase: a cloud browser the agent drives with navigate, act, observe and extract. Free: 1 browser hour and 3 concurrent sessions, without CAPTCHA solving or proxies. Paid: Developer plan from $20.
• Prices and free tiers were checked on each vendor's pricing or docs page on 2026-10-20 and change often.

Bright Data's MCP server routes every request through its Web Unlocker, which its docs say handles anti-bot detection, CAPTCHAs and proxy rotation automatically. Use it when a site blocks plain fetches. The hosted endpoint takes your API token in the URL; the local server reads API_TOKEN (the GitHub README's name; one docs page says API_KEY).

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

{
  "mcpServers": {
    "brightdata": {
      "command": "npx",
      "args": ["@brightdata/mcp"],
      "env": { "API_TOKEN": "YOUR_API_TOKEN" }
    }
  }
}

• Main tools: search_engine, scrape_as_markdown, scrape_batch, discover, scraping_browser_* (browser control) and web_data_* (structured data, such as product or profile pages).
• Optional env vars: WEB_UNLOCKER_ZONE, BROWSER_ZONE, RATE_LIMIT, GROUPS and TOOLS to limit which tools load.

• Bright Data MCP server: install for every agent

Firecrawl's server turns pages into Markdown or into JSON that follows a schema you give it, and can crawl or map a whole site. Its docs say to send the API key as a Bearer header, never in the URL, or to use the OAuth endpoint.

{
  "mcpServers": {
    "firecrawl": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": { "FIRECRAWL_API_KEY": "fc-YOUR_KEY" }
    }
  }
}

• Hosted: https://mcp.firecrawl.dev/v2/mcp with Authorization: Bearer, or https://mcp.firecrawl.dev/v2/mcp-oauth to sign in.
• Main tools: firecrawl_scrape, firecrawl_search, firecrawl_map, firecrawl_crawl, firecrawl_interact, firecrawl_agent.
• Self-hosted Firecrawl: set FIRECRAWL_API_URL.

• Firecrawl MCP server: install for every agent

Apify's server lets the agent find a ready-made scraper for a site (Google Maps, Instagram, Amazon and thousands more in the Apify Store) and run it. The hosted endpoint supports OAuth sign-in, which is the simplest setup.

claude mcp add --transport http apify https://mcp.apify.com

• Local: npx -y @apify/actors-mcp-server with APIFY_TOKEN.
• Main tools: search-actors, fetch-actor-details, call-actor, apify/rag-web-browser, search-apify-docs.

• Apify MCP server: install for every agent

• Tavily: hosted at https://mcp.tavily.com/mcp/ (API key as tavilyApiKey, a Bearer header, or OAuth); local npx -y tavily-mcp@latest with TAVILY_API_KEY. Pick it when the agent mostly needs search results, not full pages.
• Browserbase: hosted at https://mcp.browserbase.com/mcp with browserbaseApiKey; local npx @browserbasehq/mcp with BROWSERBASE_API_KEY. Pick it for multi-step flows such as logging in and clicking through pages. CAPTCHA solving and proxies are paid-plan features.

• Tavily MCP server
• Browserbase MCP server

• Sites that block bots, or structured data from big sites: Bright Data.
• Clean Markdown or JSON from docs and blogs, or a whole-site crawl: Firecrawl.
• A scraper someone already built for a specific site: Apify.
• Search answers with sources: Tavily.
• Clicking through a site like a person: Browserbase.
• Keep API keys out of committed files: use environment variables or the OAuth endpoints where the vendor offers them.

• Set up MCP servers, rules and permissions in one paste
• Add an MCP server in every AI coding agent