Top web scraping and research MCP servers

Ranked by the composite AgentRank score — a weighted blend of stars (15%), freshness (25%), issue health (25%), contributors (10%), and inbound dependents (25%). Average score across all 820 search and scraping tools is 28.4. The tools below are in the top tier.

Choosing by use case

Full research pipeline (search + extract + crawl)

tavily-ai/tavily-mcp leads the category at 81.36 because it does everything: real-time web search, content extraction from URLs, site crawling, and site mapping. 1,410 stars and 14 contributors. If your agent needs to go from a question to a structured answer — search for sources, extract the relevant content, organize it — Tavily covers the full pipeline without chaining multiple servers.

Semantic / AI-native search

exa-labs/exa-mcp-server scores 76.64 with 4,035 stars and the most recent commit of any search server (March 17). Exa's search engine is optimized for semantic similarity — finding pages that are conceptually related to your query, not just keyword-matched. The AgentRank score advantage over Firecrawl comes partly from better issue health: a 59% issue-close rate vs. Firecrawl's alarming backlog. Use Exa when keyword search returns noise and you need conceptually relevant results.

Web scraping and structured data extraction

firecrawl/firecrawl-mcp-server is the most-starred web research server in the index at 5,798 stars — but scores 73.94, slightly behind Exa, due to an open issue backlog. It's a crawl-first tool: turn any URL into clean markdown, extract structured data from pages, or batch-crawl entire sites. 21 contributors, official Firecrawl team maintenance. If your agent needs to scrape specific URLs or extract structured data from pages, Firecrawl is the right call.

Privacy-first search

brave/brave-search-mcp-server scores 70.32 with 774 stars and 17 contributors. Official Brave browser team server. It uses Brave's independent search index — no Google data, no Google tracking. Supports both web search and local search. If your workflow requires an independent search index or if data privacy is a hard requirement, this is the only major search MCP with its own non-Google index.

Pre-built scrapers at scale (Apify)

apify/apify-mcp-server takes a different approach at 64.4: one MCP server that exposes 19,000+ pre-built Apify Actors — ready-to-run scrapers for social media, e-commerce, job boards, maps, and more. 899 stars, 14 contributors, official Apify team server. If you need data from a specific platform (LinkedIn, Instagram, Amazon, Google Maps), there's almost certainly a pre-built Actor for it. Far cheaper than building a custom scraper.

Simple page fetch

jae-jae/fetcher-mcp scores 43.9 with 1,008 stars but a January 14 last commit date — the freshness penalty explains the score gap. For simple use cases (fetch a URL, return clean text), it's the lightest-weight option in this list. Playwright-powered so it handles JavaScript-rendered pages. Only 5 contributors and no recent commits means you're accepting maintenance risk.

Search vs. scraping vs. crawling

These terms get conflated. They're different operations with different right tools:

Web search

Give the agent a query, get back a list of URLs and snippets. No page content — just metadata. Best for: finding sources, discovering what exists on a topic. Tools: Tavily, Exa, Brave.

Web scraping

Fetch a specific URL and extract its content — as clean text, as markdown, or as structured data. Best for: extracting information from known pages. Tools: Firecrawl, fetcher-mcp.

Site crawling

Follow links from a starting URL to index an entire site. Best for: building knowledge bases from documentation, analyzing a full domain, bulk data extraction. Tools: Firecrawl, Tavily (partial crawl support).

Pre-built extraction

Run a specific extractor tuned for a platform (LinkedIn profiles, Amazon prices, Google Maps listings). Best for: platform-specific structured data without custom scraper development. Tool: Apify.