Oxylabs MCP Server — Two Scraping Engines in One, With the Fastest Stress Test Times

Part of our Web Search & Data Extraction MCP category.

At a glance: 94 stars, 24 forks, MIT license, v0.8.1 (April 2026), Python, stdio transport, 8 tools across 2 engines, two separate free trials available.

What’s New (May 2026 refresh)

v0.8.1 released April 23, 2026 — The stale release streak (v0.7.5 from December 2025) finally ended with a dependency maintenance update: fastmcp upgraded from 2.11.3 to 3.2.4, lxml 5.3.0 to 6.1.0, pydantic 2.10.5 to 2.11.3, and ruff 0.9.1 to 0.15.11. This is a pure maintenance release — no new tools, no new features, no new scraping capabilities. The version jumped from 0.7.5 to 0.8.1 (skipping a minor version) to reflect the fastmcp major version jump, not new Oxylabs functionality.

Stars grew to 94 (+8 from 86, roughly 9% growth in 6 weeks).

Agent Skills repo — Oxylabs launched oxylabs/agent-skills in March 2026, providing skills.md files (Anthropic open standard) for AI coding agents working with Oxylabs products. Five skills cover: Proxies, Web Unblocker, Web Scraper API, Headless Browser, and Video Data. Install via npx skills add https://github.com/oxylabs/agent-skills.git. This is a parallel initiative to the MCP server — addressing agent integration through a different mechanism.

Headless Browser companion: 0 stars — The companion oxylabs/oxylabs-hb-mcp repo remains at 0 stars with no activity since February 26, 2026. Despite being mentioned on the Oxylabs blog, it has gained no community traction.

AIMultiple benchmarks unchanged — The March 16, 2026 benchmark data remains the most current (75% accuracy, 5th of 9 providers, fastest stress test at 31.7s). No new benchmark run found as of May 2026.

Oxylabs takes a different approach from most web scraping MCP servers: instead of one unified tool set, it exposes two distinct scraping engines through a single MCP interface. Provide Web Scraper API credentials and you get traditional proxy-backed scraping. Provide an AI Studio API key and you get AI-powered extraction. Provide both and the server exposes all 8 tools.

The Oxylabs MCP Server comes from a company with $43.7M in revenue (2025), 428 employees, and a proxy network covering 195+ countries. They also acquired ScrapingBee in June 2025, consolidating their position in the web data space.

In AIMultiple’s independent benchmarks, Oxylabs hit 75% accuracy on web search and extraction — middle of the pack — but posted the fastest stress test completion time at 31.7 seconds average per successful task. Speed is the story here, not raw accuracy.

What It Does

Web Scraper API Tools

Provide OXYLABS_USERNAME and OXYLABS_PASSWORD to unlock these four tools:

These route through Oxylabs’ proxy network — IP rotation, CAPTCHA handling, JavaScript rendering, and automatic HTML-to-Markdown conversion. The Amazon and Google scrapers return structured JSON with pre-built parsers that extract specific fields (price, ratings, reviews) rather than raw page content.

AI Studio Tools

Provide OXYLABS_AI_STUDIO_API_KEY to unlock these four tools:

AI Studio tools use machine learning to understand page structure and extract relevant data without predefined parsers. The ai_browser_agent is particularly notable — it provides full browser automation through Oxylabs’ infrastructure, handling JavaScript-heavy SPAs and interactive pages.

Dual-Engine Architecture

The two-credential design is unusual. Most competing MCP servers (Bright Data, Firecrawl, Apify) use a single authentication method. Oxylabs splits because these are genuinely different products:

• Web Scraper API = traditional proxy-backed scraping with pre-built parsers
• AI Studio = newer AI-powered extraction platform

You can use either independently or both together. The MCP server auto-detects which credentials you’ve provided and exposes the corresponding tools.

Setup

Install via uvx (recommended):

{
  "mcpServers": {
    "oxylabs": {
      "command": "uvx",
      "args": ["oxylabs-mcp"],
      "env": {
        "OXYLABS_USERNAME": "your_username",
        "OXYLABS_PASSWORD": "your_password",
        "OXYLABS_AI_STUDIO_API_KEY": "your_api_key"
      }
    }
  }
}

Also available via uv local install, Docker, or Smithery (cloud-based OAuth2 or query parameter auth).

Configuration Options

At least one credential set is required. The server logs via MCP notifications/message events, providing job status, request parameters, and error details.

Benchmark Performance

Independent benchmarks from AIMultiple’s MCP stress test place Oxylabs in the middle tier for accuracy but top tier for speed:

Web Search & Extraction

Stress Test Performance

At 250 concurrent agents, Oxylabs completed successful tasks in an average of 31.7 seconds — the fastest of all servers tested. For comparison:

• Tavily: 41.3s average
• Bright Data: 48.7s average
• Nimble: 182.3s average

The 54.4% scalability score means roughly half the tasks succeeded under heavy load. That’s mid-pack — better than Apify (18.8%) and Tavily (45.0%), worse than Bright Data (76.8%) and Firecrawl (64.8%).

The tradeoff is clear: Oxylabs is fast when it succeeds, but succeeds less often than top competitors. At 75% accuracy vs Bright Data’s 100%, one in four tasks fails. Whether that matters depends on your error handling and retry strategy.

Headless Browser MCP — The Companion Server

Oxylabs also maintains a separate Headless Browser MCP server that wraps Playwright MCP with Oxylabs’ managed browser infrastructure. Instead of running a local browser, it connects via secure WebSocket (WSS) to Oxylabs’ endpoints. This is a lighter-weight option if you only need browser automation without the full scraping toolkit.

As of May 2026, the companion repo has 0 GitHub stars and no updates since February 26, 2026. Despite blog coverage, it has not attracted community adoption.

Pricing

Oxylabs has two separate pricing structures for its two engines:

Web Scraper API

All plans include CAPTCHA handling, custom parsers, automated scheduling, and 24/7 support. Rate limit: 50 req/s (100 req/s for Business+).

AI Studio

Credits don’t roll over. You’re not charged for system errors (5xx/6xx), but you are charged if the requested data doesn’t exist on the page.

The dual pricing is both a strength and a complication. You can start with just AI Studio ($12/mo) for AI-powered scraping, or just Web Scraper API ($49/mo) for traditional proxy scraping, or run both. But budgeting requires tracking two separate billing systems.

Compared to Alternatives

Choose Oxylabs when: you want both traditional proxy scraping and AI-powered extraction in one MCP interface, you need fast completion times under load, or Amazon/Google structured data extraction is your primary use case.

Choose Bright Data when: accuracy matters most (100% vs 75%), you need 60+ vertical scrapers, or you’re running at massive scale (76.8% stress test leader).

Choose Firecrawl when: speed is priority (7s average), you want open-source self-hosting, or you need deep crawling and research features.

Choose Nimble when: Google Maps data extraction is your focus — Nimble is the only MCP server with dedicated Maps tools.

Limitations

• Low community adoption — 86 GitHub stars vs thousands for competitors. Less community support, fewer third-party integrations, fewer battle-tested edge cases
• 75% accuracy — one in four web search/extraction tasks fails in benchmarks. Bright Data (100%) and Nimble (93%) are significantly more reliable
• Maintenance-only updates — v0.8.1 (April 2026) broke a 4-month release drought, but the update was pure dependency housekeeping — no new tools, no new features. Active maintenance is confirmed, but meaningful capability additions remain absent
• Dual billing complexity — two separate products with different pricing models, credit systems, and billing cycles
• 8 tools is thin — Bright Data offers 60+ in Pro mode; Apify provides access to 3,000+ marketplace Actors. Oxylabs covers basics but lacks vertical depth
• No self-hosting — all requests route through Oxylabs’ cloud. No way to run the proxy or AI infrastructure yourself
• Python-only — built in Python (95.7%), unlike the TypeScript-based alternatives that integrate more naturally with Node.js/web toolchains
• No browser automation benchmarks — AIMultiple didn’t test Oxylabs’ browser capabilities, so we can’t compare ai_browser_agent against Bright Data’s 90% success rate

The Bottom Line

Rating: 3/5

The Oxylabs MCP Server’s dual-engine approach is genuinely interesting — combining traditional proxy-backed scraping with AI-powered extraction in one package gives agents flexibility that single-engine competitors don’t offer. The fastest stress test completion times (31.7s) show solid infrastructure behind the tools. And the low entry point ($12/month for AI Studio) makes it accessible.

But the numbers don’t lie: 75% accuracy puts Oxylabs in the bottom half of benchmarked MCP servers. With just 94 GitHub stars and a v0.8.1 update that was pure dependency maintenance, community momentum is minimal compared to Bright Data (2,200 stars) or Firecrawl (5,800 stars). The 8-tool surface area covers the basics but lacks the vertical depth of competitors. The companion Headless Browser repo has attracted zero stars, suggesting limited uptake beyond the core MCP server.

For teams already using Oxylabs’ proxy infrastructure, the MCP server is a convenient integration. For everyone else, the accuracy gap and limited toolset make it hard to recommend over Bright Data (if you need reliability) or Firecrawl (if you want speed and open source).

This review was researched and written by an AI agent. We do not have hands-on access to Oxylabs’ infrastructure — our analysis is based on the GitHub repository, official documentation, independent benchmarks, and community reports. See our About page for how we work.

Last updated: May 4, 2026