HasData Review 2026: 57 MCP Tools, AI Mode Data and a Credit System Worth Reading Twice

What is HasData?

HasData is a web scraping platform that sells structured public web data through APIs. It covers two things that usually come from separate vendors: a general Web Scraping API that fetches any public URL through managed proxies, and a set of source-specific scrapers for Google search, Maps, Amazon, YouTube, TikTok, Zillow, Yelp and about 20 other sources.

The company is small and Latvian-registered, with 1,300+ customers by its own count. That matters for the buying decision, so I will come back to it.

The part I found most interesting is the AI search coverage. HasData ships a Google AI Mode API and an AI Overview endpoint that return the generated answer text along with the sources Google cited. For anyone tracking whether their brand gets mentioned in AI answers, that is the raw material, and most SERP API vendors still do not sell it.

Does HasData have an MCP server?

Yes. HasData runs an official MCP server at mcp.hasdata.com/mcp over streamable HTTP, and it exposes 57 tools. I counted them from a live `tools/list` call rather than trusting the docs, and 57 is exactly right.

Authentication accepts either an `x-api-key` header or OAuth. An unauthenticated request returns a proper 401 with a `WWW-Authenticate` challenge and a pointer to the OAuth resource metadata, which is the correct behaviour and rarer than it should be. I connected with the API key and had results back on the first call.

The tool descriptions are the best thing about it. Each one documents parameter semantics, pagination tokens and expiry windows, so an agent can pick the right call without a human explaining the API first. The AI Overview tool, for example, states that its page token is valid for four minutes.

Two caveats. The server tells you outright that its "tools are generated from generated/endpoints.json", so it is an auto-generated passthrough over the REST API rather than a designed agent surface, and tool results arrive as a raw HTTP envelope of `{url, status, json, text}` that your agent has to unwrap. One tool name also got hash-truncated by the code generator to fit the 64-character limit, arriving as `hasdata_google_serp_immersive_product_getImmersive_e29f691177`.

None of that blocks anything. It just shows the MCP layer is generated, not crafted.

What does the HasData API cover?

The Google surface is the deepest part. A single Google SERP API call came back with organic results, ads, People Also Ask, related searches, pagination and an AI Overview block, each under its own key. There are separate endpoints for News, Trends, Shopping, Images, Flights, Hotels, Events and short videos.

Beyond Google it covers Bing and DuckDuckGo search, Google Maps places, reviews, photos and posts, Amazon products, reviews and sellers, YouTube videos, channels and transcripts, TikTok, Instagram, Yelp, Yellow Pages, Zillow, Redfin, Indeed, Glassdoor, Booking, Airbnb and Shopify.

How good is the AI extraction?

Better than I expected. The Web Scraping API takes an `aiExtractRules` object where you describe the fields you want in plain English, and it returns structured JSON. I pointed it at a live pricing page and asked for plan name, monthly price and included credits as a list. It returned four clean plan objects on the first attempt, in 7.2 seconds, with no selectors written and no retries.

That is the feature I would actually pay for. Writing CSS selectors against a marketing page is work that breaks every time the vendor ships a redesign, and describing the fields instead removes that maintenance entirely.

How fast is HasData?

I timed every call rather than repeating the marketing numbers. A Google SERP API request returned in 2.7 seconds, near identical whether I called it over REST or through MCP, which tells you the MCP layer adds almost no overhead.

The Google AI Mode API is the slow one at 11.3 seconds. That is the cost of Google generating an answer rather than returning a cached page, and it came back with 20 text blocks and 8 structured references, each carrying link, title, source and citation index. For citation tracking that payload is worth the wait, but do not put it in a synchronous user-facing path.

Concurrency held up. I fired 20 simultaneous Google SERP Light requests on a 50-concurrency account and got 20 out of 20 back successfully, 200 organic results total, in 4.7 seconds wall clock. Individual requests ranged from 0.92 to 4.64 seconds. No 429s, no throttling, no silent truncation.

My whole test session, roughly 30 calls including that concurrency burst, cost 195 credits out of 20,000.

Where does the AI Overview data get awkward?

One thing will bite you if you build against it blind. The AI Overview block comes back in two different shapes depending on the query. For "what is generative engine optimization" the full answer text arrived inline inside the SERP response. For "best seo api" the same field held only a `pageToken`, valid for four minutes, which needs a second billed call to resolve into the actual answer.

Both shapes are legitimate and the docs describe the token flow. But your code has to branch on which one it got, and if you only ever tested the inline case you will ship a parser that silently returns empty answers. Budget for two calls per keyword when you are costing out AI Overview tracking.

HasData gave me an account with 20,000 credits and 50 concurrency, so I ran the things a buyer would actually check rather than reading the docs back to you.

I connected to the MCP server with an API key and counted the tools it advertises: 57, which matches what HasData claims. I pulled a live Google SERP and got results back in 2.7 seconds, then did the same call over REST to see whether MCP added overhead. It did not. I asked the AI Mode endpoint a question and it took 11.3 seconds to return 20 blocks of answer text and 8 cited sources. I pointed the scraper at a live pricing page with plain-English extraction rules and it returned four clean plan objects in 7.2 seconds, first try, no selectors.

Then I fired 20 requests at once to see whether the concurrency number is real. All 20 came back, 200 organic results, 4.7 seconds wall clock, no rate limiting. Afterwards I checked the billing against the usage endpoint, which is where I found the numbers disagreeing. The whole session cost 195 of 20,000 credits.

How much does HasData cost?

HasData starts free and runs $49, $99 and from $249 per month on monthly billing, with a custom Enterprise tier above that. Annual billing is sold as two months free, so Startup drops to about $40.83 per month ($490 a year) and Business to $82.50 per month ($990 a year).

The free tier is a real one: 1,000 credits a month, no credit card, plus a 30-day trial that opens up the paid features. The catch is 1 concurrent request and results deleted after 7 days, so it is sized for checking coverage against your own queries, not for running anything.

What is a HasData credit worth?

This is the number everyone gets wrong, including me before this week. Credits are not requests. A Web Scraping API call costs 1 credit, a Google Maps call costs 5, and a Google SERP API call costs 10. So a per-credit rate is never a per-request rate, and quoting one as the other understates the price of SERP work by roughly ten times.

HasData publishes the honest version of this as a per-1,000-requests table, broken out by product. For Google SERP API data specifically:

• Free: $0, capped at 100 SERP requests a month
• Startup ($49/mo): $2.45 per 1,000 requests, 20,000 included
• Business ($99/mo): $0.99 per 1,000 requests, 100,000 included
• Growth (from $249/mo): $0.83 per 1,000 requests, 300,000 included

The general Web Scraping API is far cheaper, because it is a far cheaper thing to do: $0.24 per 1,000 on Startup, falling to $0.08 per 1,000 on Growth. Google SERP Light, which drops the rich blocks and returns organic results only, sits in between at $1.23 per 1,000 on Startup.

So the read is straightforward. HasData is cheap for general web scraping and mid-priced for SERP data. If you are comparing it against SerpApi or ValueSERP on SERP volume alone, $2.45 per 1,000 at the entry tier is not a bargain. If you need both SERP data and general scraping from one vendor, the combined bill usually beats buying them separately.

One genuinely good billing detail: only successful requests consume credits. Failed and invalid requests are not charged. That is documented and repeated in the app, and it removes the usual argument about paying for blocked pages.

A small inconsistency worth knowing before you do the maths: the in-app annual toggle is labelled "-20%" but the actual discount is two months free, about 16.7%. The prices it shows are correct. The badge is just optimistic.