Top web scraping & browser 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 1340 web/browser tools is 35.8. The tools below are the category leaders.

Choosing by use case

General browser automation

microsoft/playwright-mcp is the dominant browser automation MCP at 85.0 — the highest score in the web/browser category and one of the highest across the entire index. The key design decision: it uses Playwright's accessibility tree to interact with pages rather than pixel-level screenshots. This means agents can click, fill forms, and navigate without requiring vision capabilities. 52 contributors, Microsoft-maintained, updated weekly. The default choice for any agent that needs to interact with web interfaces.

Browser inspection and DevTools

ChromeDevTools/devtools-mcp scores 69.2 with 3,807 stars. Where Playwright-MCP is optimized for automation (click, fill, navigate), the Chrome DevTools MCP exposes the full DevTools Protocol — network inspection, JavaScript execution, performance profiling, DOM manipulation. Use it when you need an agent to debug, inspect, or profile a live browser session rather than just automate it. Maintained by the Chrome DevTools team.

Pre-built scraping at scale (Apify)

apify/apify-mcp-server at 64.4 solves a different problem than Playwright or DevTools. Instead of building a scraper from scratch, it gives agents access to 19,000+ pre-built Actors — scrapers for Instagram, Amazon, Google Maps, LinkedIn, and essentially every major site. When you need data from a well-known source and don't want to build or maintain a custom scraper, Apify's MCP turns that into a one-line agent call.

Cloud browser infrastructure (Browserbase)

browserbase/mcp-server-browserbase scores 47.9 with 3,189 stars. The distinction from Playwright-MCP: Browserbase runs the browser in the cloud. No local browser required. It handles anti-bot fingerprinting, residential proxy rotation, and CAPTCHA solving infrastructure. The Stagehand layer on top adds AI-native selectors — instruct by description, not CSS path. For production scraping at scale, this is the operationally cleanest path.

Full-site crawling (Firecrawl)

firecrawl/firecrawl-mcp-server at 47.7 focuses on extraction quality. It crawls an entire site or single URL and returns clean markdown — stripping ads, navigation, and boilerplate. Handles JavaScript rendering. Supports structured data extraction with schemas. With 5,759 stars, it has the strongest community signal in the scraping subcategory. Use it when you need readable content output rather than raw HTML or browser control.

Simple page fetching

jae-jae/fetcher-mcp at 43.9 is the lightweight option with 1,008 stars. It runs a Playwright headless browser under the hood but exposes a simple fetch interface — give it a URL, get back clean content. No orchestration layer, no cloud infrastructure. Use it when you need occasional page fetches in an agent workflow without the overhead of the full Playwright-MCP setup.

Puppeteer-based automation

docentovich/chrometools-mcp at 39.9 is the Puppeteer alternative to Playwright-MCP. The key difference: persistent session support. Playwright-MCP creates fresh browser contexts by default; chrometools-mcp maintains state across agent calls. Useful for multi-step workflows where the agent needs to stay logged in, navigate through paginated results, or maintain form state across turns.

What to watch for

Playwright vs Puppeteer: the practical choice

Both Playwright and Puppeteer control Chromium via the DevTools Protocol. Playwright also supports Firefox and WebKit and has better built-in test infrastructure. For MCP use cases in 2026, Playwright is the clear default — microsoft/playwright-mcp at 28,849 stars versus fragmented Puppeteer MCP options. The exception: if your workflow depends on specific Puppeteer plugins or existing Puppeteer scripts, the chrometools-mcp preserves that compatibility.

Build vs buy for scraping

The MCP ecosystem offers a clear ladder: for one-off page reads, use fetcher-mcp. For structured data extraction from known sites, Apify saves weeks of scraper maintenance. For custom, complex workflows against arbitrary sites, Playwright-MCP gives you full control. For production at scale without infrastructure headaches, Browserbase handles the hard parts. Match the tool to the operational complexity you're willing to manage.

Anti-bot and legal considerations

Web scraping at agent speeds triggers anti-bot protection on most commercial sites. Playwright-MCP and DevTools-MCP expose raw browser automation without anti-detection measures — expect CAPTCHAs and blocks on anything with aggressive bot detection. Browserbase includes anti-bot infrastructure by design. Regardless of tooling, scraping must comply with a site's Terms of Service and applicable law; the presence of an MCP server does not change the legal analysis.