ScraperAPI MCP Server: What Is It, How to Set It Up, Which Plan Should You Choose, and Is It Actually Worth It? (Complete Guide with Pricing Breakdown, Supported Platforms, and Free Trial Info)
Ever sat there wishing your AI assistant could just… go grab data from the web on its own? Like, you're in Claude, you want to pull competitor pricing from a dozen product pages, and instead of copy-pasting like it's 2012, you just tell it what to grab — and it does.
That's the whole pitch of the ScraperAPI MCP server. And honestly, after digging into this thing properly, it's more interesting than it sounds on the surface.
Let me walk you through what the ScraperAPI MCP server actually is, how to get it running, which plan makes sense for what you're doing, and where it fits in the broader landscape of web scraping tools for AI in 2026.
Before anything else — MCP stands for Model Context Protocol, the open standard Anthropic introduced that lets AI models talk to external tools in real time. Think of it as giving your LLM hands. Instead of being stuck with whatever was in its training data, it can reach out, pull live information, and bring it back into the conversation.
The ScraperAPI MCP server is ScraperAPI's implementation of this idea. It connects your LLM — Claude Desktop, Claude Code, Cursor, Windsurf, VS Code with GitHub Copilot, Cline, you name it — directly to ScraperAPI's scraping infrastructure. That means your AI can route web requests through ScraperAPI's proxy pool, handle JavaScript rendering, bypass anti-bot systems, and return structured data — all triggered by a plain-language prompt.
No API calls written by hand. No code debugging at 11pm. You write a prompt, the MCP server translates it into a ScraperAPI request, and the result comes back ready for your AI to process.
ScraperAPI itself has been around since 2018 and currently processes over 36 billion API requests per month for clients including Deloitte, Sony, and Alibaba. It sits on a rotating pool of 40 million+ IPs across 50+ countries. The MCP server is the layer that makes all of that accessible through natural language.
One thing worth knowing upfront — the ScraperAPI MCP server isn't one-size-fits-all in how you deploy it. There are two modes:
Remote server (no local setup required) This is the easiest path. Your AI client connects to ScraperAPI's hosted MCP endpoint. No installation, no Python environment, no Docker — it just works. Claude Desktop's hosted remote MCP support handles this natively.
Local server (self-hosted via Python or Docker) If you want more control — maybe you're building automated pipelines, integrating into a custom workflow, or have security requirements around routing traffic locally — you can run the server yourself. Both Python and Docker are supported.
bash
pip install scraperapi-mcp-server
docker pull scraperapi/mcp-server
In both cases, when your AI client needs to make a scraping request, the MCP server starts automatically, routes it through ScraperAPI's infrastructure, and delivers the response back.
This expanded significantly in early 2026. Here's what's currently on the table:
Scraping tools available through MCP:
- Standard web page fetching (with full ScraperAPI parameter support — geotargeting, premium proxies, JS rendering, etc.)
- Structured Data Endpoints (SDEs) — Google, Amazon, Walmart, eBay, and Redfin, accessible via natural language prompt
- Full web crawling through MCP protocol (not just single pages — whole site crawls)
Platform support:
- Claude Desktop (hosted remote MCP)
- Claude Code (native MCP tool execution)
- Cursor
- Windsurf
- Cline
- VS Code + GitHub Copilot
- Local Python and Docker deployments
Language model compatibility:
- Anthropic Claude (primary)
- OpenAI compatibility ready
- Google Gemini compatibility ready
There's also a ScraperAPI MCP Skill — an intelligent agent layer you add on top that auto-detects the task type (is this a scrape? a search? a crawl? a structured data query?), picks the right tool, and applies best practices around cost efficiency and pagination. You add it like this:
bash npx skills add scraperapi/scraperapi-skills
Practical upshot: you don't have to think about which ScraperAPI tool to call for a given request. The skill figures that out.
Here's the practical flow for most people:
Step 1: Get a ScraperAPI account and API key
👉 Start your free trial — 5,000 credits, no credit card required
You'll need an API key from your ScraperAPI dashboard. This is what authenticates your MCP server's requests.
Step 2: Choose your deployment mode
For most people testing this out, the remote server is the right call. For Claude Desktop, you add ScraperAPI to your MCP client config. For local Python deployment, you install the package and configure your API key.
Step 3 (optional): Add the ScraperAPI MCP Skill
bash npx skills add scraperapi/scraperapi-skills
This gives your AI agent better decision-making around which tool to use for a given task.
Step 4: Start prompting
Once the server is connected, you can do things like:
- "Scrape the pricing page at [URL] and give me a table of all plans and prices"
- "Search Google for the top 10 results for [query] and summarize what each page is about"
- "Crawl [website] and extract all product names and SKUs"
- "Pull the Amazon product page for ASIN B09XXXXX and return the price, rating, and top 3 reviews"
The MCP server handles the proxy rotation, JavaScript rendering decisions, and anti-bot bypass automatically. You just describe what you want.
Here's where I need to be honest with you about something most articles gloss over: ScraperAPI's credit system has multipliers that make the advertised numbers somewhat misleading at face value.
The headline says "100,000 credits" — but 1 credit ≠ 1 request in many cases. Here's how it actually works:
| Target Domain | Base Credits per Request |
|---|---|
| Standard web pages (blogs, news, etc.) | 1 credit |
| E-commerce sites (Amazon, Walmart, eBay) | 5 credits |
| Search engines (Google, Bing) | 25 credits |
| Social networks (LinkedIn) | 30 credits |
And feature parameters stack on top:
| Feature | Additional Credits |
|---|---|
JavaScript rendering (render=true) |
+10 |
Premium proxies (premium=true) |
+10 |
Screenshots (screenshot=true) |
+10 |
Ultra-premium proxies (ultra_premium=true) |
+30 |
| Premium proxy + JS rendering (combined) | +25 (not +20) |
| Ultra-premium + JS rendering (combined) | +75 (not +40) |
That combined pricing matters. The non-linear stacking is where people get surprised. An Amazon scrape with JavaScript rendering on ultra-premium proxies costs 5 (domain) + 75 (combined feature) = 80 credits per request. Your "100,000 credit" Hobby plan gets you about 1,250 of those. Know this before you start.
On the positive side: you're only charged for successful requests (200 and 404 status codes). Failed scrapes don't eat credits.
| Plan | Monthly Price | Annual Price (10% off) | API Credits / Month | Concurrent Threads | Geotargeting | Link |
|---|---|---|---|---|---|---|
| Free Trial | $0 (7 days) | — | 5,000 (one-time) | 5 | Limited | Start Free Trial |
| Free Tier | $0 ongoing | — | 1,000 | 5 | Limited | Get Free Tier |
| Hobby | $49/month | $44.10/month | 100,000 | 20 | US & EU only | Get Hobby Plan |
| Startup | $149/month | $134.10/month | 1,000,000 | 50 | US & EU only | Get Startup Plan |
| Business | $299/month | $269.10/month | 3,000,000 | 100 | Global (50+ countries) | Get Business Plan |
| Scaling | $475/month | $427.50/month | 5,000,000 | 200 | Global | Get Scaling Plan |
| Professional | $975/month | $877.50/month | 10,500,000 | 300 | Global | Get Professional Plan |
| Advanced | $1,975/month | $1,777.50/month | 21,500,000 | 500 | Global | Get Advanced Plan |
| Enterprise | Custom | Custom | 22,000,000+ | 500+ | Global | Contact Sales |
A few non-obvious things from this table:
- Geotargeting beyond US and EU is gated to Business and above. If you need country-level proxy targeting for Germany, Japan, Brazil — anything outside the US/EU cluster — you're starting at $299/month.
- Pay-as-you-go overflow only kicks in on Scaling and above. On Hobby, Startup, and Business, running out of credits mid-month means you're cut off until the next billing cycle. No overflow option.
- Credits don't roll over. Whatever you don't use resets at renewal.
- Professional and Advanced are the new Growth tier plans (introduced May 2026) designed for teams that need higher volume without going full enterprise negotiation. Professional includes a 250K bonus credit offer; Advanced includes 500K bonus credits as a limited-time offer.
Rather than a generic "it depends," here's how to actually think about this:
Pick the Free Trial first, no matter what. You get 5,000 credits, no card required. Point it at your actual scraping targets — not a toy example — and watch what happens to your credit consumption. This is the only real way to know what you're actually signing up for.
Pick Hobby ($49/month) if you're building something personal, running a side project, or just testing the MCP server integration in Claude. 100,000 credits handles a lot of standard page scraping, though it shrinks fast if Amazon or Google is in the mix.
Pick Startup ($149/month) if you've moved past casual use and you're running consistent scraping jobs — a small SaaS that needs regular data refreshes, or an agency with a handful of clients. The jump to 1 million credits and 50 concurrent threads is meaningful.
Pick Business ($299/month) if you need global geotargeting, unlimited analytics history, or you're running production infrastructure that other parts of your business depend on. This is also the first tier with 100 concurrent threads, which matters for parallelized MCP-driven scraping agents.
Pick Scaling or above if volume is the main constraint and you want pay-as-you-go overflow so you're never hard-stopped mid-month.
The MCP server inherits ScraperAPI's underlying strengths and weaknesses. From independent benchmarks (Scrapeway, April 2026):
| Target | Success Rate | Notes |
|---|---|---|
| Zillow | ~100% | One of the strongest domains |
| Amazon | ~98% | Structured Data Endpoint adds parsed JSON |
| Etsy | ~99% | Reliable |
| ~95% | Expensive at 30 credits/request | |
| Walmart | ~93% | Good with SDE |
| 0% | Complete failure | |
| Twitter/X | 0% | Complete failure |
| Booking.com | 0% | Complete failure |
For MCP-driven workflows, this shapes what kinds of prompts will actually return results. Asking your AI to "pull data from Amazon product pages" works beautifully. Asking it to "scrape Instagram for brand mentions" — no, that's not happening.
ScraperAPI also explicitly prohibits scraping data behind login walls. So if your MCP workflow involves "log into this platform and pull my account data," that's out of scope. The MCP server works on publicly accessible pages.
Fair question: if you're a developer who can write API calls, why use the MCP server at all?
A few genuine reasons:
Speed of iteration. Instead of writing a Python script to hit the ScraperAPI endpoint, loop through URLs, handle errors, and parse output — you describe the task in natural language. For prototyping or ad-hoc data needs, this is meaningfully faster.
Agentic workflows. The MCP server lets your AI take scraping as one step in a larger multi-step agent workflow. "Pull competitor pricing from these 20 URLs, compare against our current prices, identify where we're underpriced by more than 15%, and format the findings as a CSV" — that's an agent task, not a script. The MCP integration is what makes it possible without a human in the loop at each step.
Structured Data Endpoints through natural language. Calling the Amazon SDE through the MCP server with a natural language prompt — vs. manually constructing the API call, handling pagination, and writing the JSON parser — is a real time savings.
When does direct API usage make more sense? High-volume production pipelines, situations where you need precise control over retry logic and headers, or when cost efficiency at scale is the primary driver. At 10 million requests a month, writing optimized API calls beats the overhead of MCP routing.
Across G2 (4.4/5), Capterra (4.6/5), and Trustpilot (4.5/5), the pattern is consistent:
"Super easy to set up. You can start scraping in minutes." — Latenode community review
"Breakdown of credit costs can be confusing." — John S., Founder, Capterra
The praise concentrates around documentation quality, easy initial integration (dropping ScraperAPI in as a proxy replacement takes minutes), and support responsiveness. The criticism concentrates around the credit multiplier system catching people off-guard and performance variability on difficult targets.
One Reddit thread documented a case where a user was quoted pricing based on 1 credit/Amazon request, paid for a plan, then discovered a 5x multiplier applied retroactively to their actual usage — creating an 80% shortfall in expected capacity. This isn't a scam, it's a documentation clarity issue, but it's a real thing that happens. The lesson: test your specific targets during the free trial before committing to a paid plan size.
Depends entirely on what you're building.
If you're a developer integrating scraping into AI agent workflows — Claude Desktop, Cursor, Claude Code — and your targets are mainstream sites (e-commerce, search, real estate), the ScraperAPI MCP server is one of the most straightforward ways to add live web data access to those agents. The free trial costs nothing to test.
If you're scraping social media, login-gated content, or highly protected sites that require complex browser automation — the tool won't get you there, regardless of which plan you're on.
If you're a non-technical user who wants to extract data from websites without writing code, the MCP server isn't really designed for you — it assumes you're connecting it to an LLM client like Claude Desktop or a coding tool like Cursor.
For what it's built for — AI-driven, prompt-based web scraping at scale on publicly accessible pages — it's a genuinely useful piece of infrastructure. The March 2026 expansion added structured data endpoints, full crawling, and broader platform support. The May 2026 Growth plans added higher-volume tiers that didn't require an enterprise sales conversation. The trajectory is clearly toward making this the default layer between AI agents and live web data.
The right move is the same as always with usage-based tools: test it with your real targets first.
👉 Start the free trial — 5,000 credits, no credit card required
Does the ScraperAPI MCP server require a paid plan? No — you can test it on the free trial (5,000 credits, 7 days, no card) or the ongoing free tier (1,000 credits/month). Paid plans unlock higher volumes, more concurrent threads, and global geotargeting.
Which AI tools work with the ScraperAPI MCP server? Claude Desktop, Claude Code, Cursor, Windsurf, Cline, VS Code with GitHub Copilot, and local Python/Docker setups. OpenAI and Gemini compatibility are also in progress.
Is there a discount for annual billing? Yes — all paid plans get 10% off when billed annually. No coupon code needed; the discount is applied automatically at checkout.
Can the MCP server scrape sites behind a login? No. ScraperAPI explicitly prohibits scraping login-gated content. The MCP server works on publicly accessible pages only.
What happens if I run out of credits mid-month? On Hobby, Startup, and Business plans, you're cut off until the next billing cycle (or you upgrade). On Scaling and above, pay-as-you-go overflow keeps you running at a fixed per-credit rate.
How do I update the local MCP server when a new version releases? bash pip install --upgrade scraperapi-mcp-server