28
36 Comments

I couldn't afford SerpApi ($10/1k) for my side project, so I built a $0.56 alternative.

Hi Indie Hackers! šŸ‘‹

I've been working as an Algorithm Engineer since 2024. Back then, I was building complex AI Agent workflows, and I hit a wall: The Knowledge Cutoff.

I relied on SerpApi for nearly a year in my daily work, but the pricing ($0.01 per search) was unsustainable for an indie budget. If my agent ran a loop and searched 50 times to debug, that was $0.50 gone in seconds. šŸ’ø

The breaking point came in Feb 2025.

I realized I couldn't keep paying these enterprise rates. So, I decided to scratch my own itch.

I've spent the time since then building and refining SearchCans – a dual-engine API specifically optimized for LLMs.

Here is what makes it different:

1. Dual Output: It returns both raw HTML (for debugging) and clean Markdown (specifically for RAG/LLMs).

2. No "HTML Junk": I added a logic to filter out ads/navbars so you don't waste tokens.

3. The Price: I managed to get the cost down to $0.56 per 1,000 requests (vs ~$10 industry standard).

The Tech Stack:

It's built with Python (Flask) + Puppeteer for rendering dynamic JS pages.

I launched it on RapidAPI to handle the key management and billing securely.

You can test it here:

šŸ‘‰ Get API Key (RapidAPI): https://rapidapi.com/hspeiyunhao/api/searchcans-google-bing-search-web-scraper

🌐 Official Website: https://www.searchcans.com

I'd love to hear your feedback on the JSON structure. Is the markdown output clean enough for your agents?

Cheers!

posted toAvatar for product SearchCans
SearchCans
  1. 2

    Wait, did I read that right? It's credits-based (pay-as-you-go) and they last for 6 months?

    That's a refreshing business model in a world full of aggressive monthly subscriptions. It’s perfect for indie devs who don't need a heavy monthly plan but want something reliable on standby. Definitely going to top up some credits just to have it in my toolkit.

    1. 1

      100% agree. Not to drift too much from the subject of the post, but I think this detail (as many other details that are not core to any one app) matters. Call it trust building, call it one more way to get the prospect closer to pulling the trigger, whatever the case I feel there are so many patterns we repeat blindly on the side of our products without realizing the world has changed and maybe one should explore innovation. This is especially true with subscription fatigue, IMO

    2. 1

      Exactly! To be honest, I haven't treated SearchCans as a profit-driven business so far. Maybe someday in the future things might change—as I’m considering hiring staff to help manage the SaaS, which will bring extra costs. But for now, my priority is to stay true to my original vision: keeping it the most cost-effective and reliable tool for the developer community.

  2. 1

    this resonates. been running into the same token cost problem with my agents - scraping full pages and feeding them to LLMs burns through credits fast.

    the clean markdown output is what i care about most tbh. most scrapers return so much garbage that you end up spending more on context tokens than the actual scraping.

    curious about a few things:

    1. how do you handle pages with heavy JS rendering? some SPAs take forever to hydrate

    2. 2. any rate limiting on your end or is it purely usage-based?

    the $0.56/1k vs $10/1k is a massive difference for anyone running agents at scale. definitely gonna check this out.

    1. 1

      Thanks for the support! You hit the nail on the head—context tokens are where the real costs hide, and that’s exactly why I focused on the Markdown output.

      To answer your questions:

      1. JS Rendering & Latency: You're right. For heavy SPAs, some pages can take 10+ seconds to fully hydrate. In my view, reliability is far more important than raw speed in these cases. I’d rather wait a few extra seconds for a successful, clean render than get a fast "403" or a blank page. We use optimized Puppeteer clusters to handle this, though the target site's complexity is the main variable.

      2. Rate Limiting: Currently, I don't have strict rate limits—it's purely usage-based. My current client base consists of 3 large companies, some RAG developers, and SEOers, so we manage the load fine. While I don't have the massive server farms of giants like SerpApi or BrightData yet, I’ve built the infra to be highly scalable. I'm adding new nodes as the user base grows to stay ahead of demand.

      The $0.56 pricing is my way of making this sustainable for fellow indie devs. Would love to hear your feedback once you've put it through its paces!

  3. 1

    This is a great example of scratching your own itch

    1. 1

      Thanks—honestly, that "itch" was getting way too expensive to scratch with the existing tools! šŸ˜…

  4. 1

    Been there — SerpApi costs add up fast once agents start looping. Clean markdown + raw HTML is a big win for LLM workflows. $0.56/1k is wild. Curious how it holds up on JS-heavy pages.

  5. 1

    Looks like SearchCans is another cool indie hacker-built tool/product. Appreciate everyone sharing use cases and feedback here, keep iterating and refining based on what users actually need!

    1. 1

      Thank you for the encouragement! You're 100% right—iterating based on real-world use cases is my top priority.

      The idea for SearchCans actually came from my own struggles as an engineer back in 2025, and seeing others find value in it now is incredibly motivating. I’m already taking notes from the feedback in this thread (especially regarding infrastructure scaling and JS rendering) to make the tool even more robust.

      More refinements are coming soon. Thanks for having my back!

  6. 1

    I’ve been there - those API costs add up fast. Out of curiosity, was this mainly slowing down customer discovery, or more of a cost annoyance while you were still testing demand?

    1. 2

      That’s a great question. For me, it was actually a mix of both, but mostly about "sustainable experimentation."

      While I was testing demand in early 2025, the $0.01/request cost made every bug and every failed agent loop feel like a personal financial loss. It creates a "fear of failure" that actually slows down the development and discovery process. You stop experimenting because you're constantly watching the meter run.

      By building SearchCans and bringing the cost down by 90%+, I removed that mental barrier. Now I (and other devs) can run thousands of tests to truly refine the product without worrying about a massive bill at the end of the month. It's about giving indie hackers the freedom to fail fast and iterate until they find that "aha!" moment.

      1. 1

        That resonates. Per-request costs quietly tax iteration speed, not just margin.
        We’ve seen teams move faster once variable costs are flattened during the experimentation phase.

        Now that cost pressure is off, what’s the next bottleneck for you: getting the right users in early, converting usage into paid adoption, or clarifying positioning so people immediately ā€œget itā€?

  7. 1

    The pricing model resonates. When you're building AI agent workflows, per-request costs compound fast — especially during development when you're iterating and debugging constantly.

    We ran into similar challenges building API infrastructure. The insight about returning clean Markdown vs raw HTML is smart — token efficiency matters when you're chaining LLM calls, and filtering out nav/ad junk before it hits the model is where the real savings are.

    Question: How are you handling rate limiting on your end? Curious if you've seen patterns in how developers integrate this into agentic loops where a single user action might trigger dozens of searches.

    1. 1

      That’s a very practical question! To be completely transparent: currently, I don’t impose a hard per-account QPS limit, but the system architecture incorporates circuit breakers and queuing mechanisms based on the available worker clusters.

      Right now, since I'm a solo founder, the physical cluster size is relatively small (typically supporting around 3-8 concurrent requests smoothly). However, the entire backend is built for infinite horizontal scaling. I can spin up additional clusters almost instantly as user demand grows.Ā 

      My philosophy is to provide a rock-solid, reliable connection rather than over-promising on throughput I can't guarantee yet. If your specific workflow requires higher concurrency, just let me know—I can easily scale up the infrastructure to accommodate your needs!

  8. 1

    I don't know where you got those pricing you are comparing to from, because I've been paying for Scrapping Bee for 3 years already and they are 3 times cheaper than your tool, and they are VERY established already in this field.

    Am I missing something?

    1. 1

      I totally get why it looks cheaper at first glance—I used to think the same way until I looked at the actual "cost per action" for my specific use cases.

      Let's do the actual math for an AI Agent workflow:

      1. The "Hidden" Multiplier:

      * Google SERP: They charge 20 credits per request.

      * JS Rendering: They charge 5 credits per request. (And let's be honest, in 2026, pure HTML pages are rare. Almost everything needs JS).

      2. The Real Cost (assuming their $49/100k credits plan):

      * SERP: 100k credits / 20 = 5,000 requests. Cost = ~$9.80 / 1k requests.

      * JS/Reader: 100k credits / 5 = 20,000 requests. Cost = ~$2.45 / 1k requests.

      * SearchCans: Flat rate of ~$0.56 / 1k requests for both.

      3. The "Use it or Lose it" Trap:

      * Their credits expire at the end of the month. If you don't use them, you lose money.

      SearchCans credits valid for *6 months**.

      So, if you are just scraping static Wikipedia pages, they are great. But for SERP and modern JS-heavy sites, SearchCans is mathematically ~4x to 15x cheaper.

      ...Actually, now that I write this math out loud, I realize I might have priced it TOO low. šŸ˜… I might need to raise the prices sooner than I planned. Better grab the credits while they last!

    2. 1

      I have conducted a comprehensive analysis of the two platforms, and in fact, their pricing calculation models differ significantly. ScrapingBee has a starting price of $49 per month – this fee is mandatory even if you only use the platform a few times in a given month. In contrast, Searchcans features a 6-month validity period for its credits. Given that many AI developers are individual users, this point-based model is far more user-friendly for this group; furthermore, this model is also gradually replacing the traditional subscription model.

  9. 1

    $0.56 vs $10 per 1K is a no-brainer for anyone running agents in loops.

    The markdown output for RAG is smart... most scraping APIs give you raw HTML and you waste tokens cleaning it up yourself. Filtering ads/navbars before it hits the LLM is the actual value here.

    What's the latency like compared to SerpApi?

    1. 1

      To be honest, comparing a solo-developer project to a giant like SerpApi on day one is tough!Ā 

      Full transparency: We had a massive traffic spike last night that actually tested our limits and caused some availability issues. It was a wake-up call.Ā 

      The main difference right now is that SerpApi has years of infrastructure behind them, while SearchCans is currently in its "rapid scaling" phase. I’m currently adding more nodes and optimizing our Puppeteer clusters to ensure we can handle this new level of demand.Ā 

      Our goal isn't to be a "SerpApi clone," but to be the most reliable, RAG-optimized alternative at 1/20th of the cost. If you can tolerate the occasional "growing pains" of a new tool, the cost savings and clean Markdown output are where we really shine.

      1. 1

        For sure man.. Most scraping APIs are priced for enterprise... $10 per 1K requests makes sense when you're a big company running occasional batch jobs. But for solo founders building AI agents that make hundreds of requests per session? Those costs kill your margins fast.

        1. 1

          100%. That's exactly the math that killed my previous project.

          When an agent needs to do a "Chain of Thought" loop and fires off 5-10 searches just to answer ONE user query, the unit economics completely break at $10/1k. You end up paying more for the search data than you do for the LLM tokens.

          I wanted to price it so we can actually afford to let these agents "think" and explore without destroying our margins.

  10. 1

    Really cool launch! It’s awesome to see a SERP + Reader API built specifically for AI developers and solo founders, especially with pricing that actually makes sense for high-volume usage. The pay-as-you-go model and lack of rate limits make it way easier to integrate real-time search into RAG pipelines or AI agents without worrying about big bills every month, and the dual APIs (search results plus URL-to-Markdown) look super useful for building everything from custom search tools to autonomous research workflows. Been looking for something this affordable and flexible, nice work!

  11. 1

    This hits close to home. ~

    Search costs have a way of sneaking up on you, especially once you start looping or debugging agents.

    I ran into the same thing where the per-query price felt harmless… right up until it was happening dozens of times per run. Suddenly I was optimizing prompts just to save a few cents, which felt completely backwards.

    The frame that helped me was pretty simple: (exploration should be cheap, production can be optimized later, debugging shouldn’t feel like you’re lighting money on fire).

    Filtering out HTML junk early is underrated too. Tokens burned on nav bars and ads add zero signal, but they quietly wreck your budget.

    One practical thing I’m curious about: did you notice whether users cared more about clean markdown or predictable structure? In my case, consistency mattered way more than prettiness once things were automated.

    Also curious—did cost pressure show up first during debugging, or only once things were live?

  12. 1

    perfect timing for your thesis. one less API to manage is always better.

  13. 1

    This is exactly the kind of comparison research that takes hours to do properly. I've seen so many developers spend days evaluating API options (pricing models, rate limits, output formats) when they just want to build their agent.

    For anyone else stuck comparing similar services (search APIs, LLM providers, database options, etc.), I offer a Technical Research service that handles the deep dive for you. Send me what you're trying to decide between and I'll send back a proper comparison.

    SearchCans looks like a solid option for cost-conscious AI workflows. Congrats on shipping!

  14. 1

    Love how you turned a personal pain point into a lean, cost-effective solution. Dual output for HTML + clean Markdown is exactly what LLM workflows need, and $0.56/1k requests is impressively accessible for indie projects.

    1. 1

      Thank you, Sonu! Coming from a positioning expert, that means a lot.Ā 

      You’re absolutely right—as a developer, I realized that "feature volume" doesn't matter if the core data is too expensive or too messy to use. My goal with SearchCans was to strip away the noise and provide the "buyer clarity" I wished I had: a predictable cost and a format (Markdown) that LLMs actually understand without extra processing.Ā 

      I’m still navigating the "growing pains" of scaling (had a wild traffic spike last day!), but hearing that the positioning resonates with a pro like you gives me a lot of confidence. Thanks for the encouragement!

  15. 1

    This is a really interesting take on search for LLMs. The dual output (raw HTML + markdown) is especially useful for debugging agent behavior.

    I’m curious how you’re handling edge cases like dynamic paywalls or heavy client-side rendering beyond Puppeteer defaults. Also, have you tested how stable the markdown output is across different site structures for long-running RAG pipelines?

    1. 1

      Man, tell me about it. Being able to diff the raw HTML vs the Markdown saved my sanity so many times while debugging.

      On the technical bits:

      For CSR, honestly, the standard Puppeteer networkidle is a lie. I stopped trusting it months ago. I had to build custom listeners for DOM mutations to actually catch when the content drops, which solves most of those "infinite loading" issues.

      For Paywalls, I don't touch server-side blocks (not worth the legal trouble), but for those annoying JS overlays or "sign up to read" popups? Yeah, the engine is pretty aggressive about nuking those to get to the text.

      And regarding stability, that's the whole point. I try to strip 99% of the noise (navs, footers, sidebars). So even if a site redesigns their layout, your embeddings shouldn't break because the core text usually stays the same.

      Shoot me a DM if you have any specific URLs that are breaking it, I'm always looking for new edge cases to fix.

  16. 1

    Congrats on the launch! Or... wait, looking at the timeline, seems like it's been live for a bit? šŸ˜‚

    This actually pops up at the perfect time. I'm currently working on my Master's thesis focused on Deep Research agents. I was just about to grab a separate Reader API key from Jina, but this looks like a much cleaner solution.

    Definitely giving it a spin—not having to manage multiple keys for search and extraction is a huge plus for my setup.

    1. 1

      That’s a great coincidence!

      I actually used these exact APIs for a Deep Research project I built for a team recently, and they performed quite well.Ā Please give them a try for your thesis. If you run into any issues or need specific optimizations for your research, just let me know. I’ll get back to you as soon as I see your message!

  17. 1

    I believe Searchcans is an excellent platform, a high-quality data infrastructure tailored for AI Agent and RAG development that boasts prominent core advantages and aligns with the practical needs of engineers. Its dual-engine architecture directly addresses the pain points in development: the SERP Engine fetches real-time search results from Google and Bing, the Reader Engine converts URLs into clean Markdown formatted for LLMs, and the bypass mode achieves a 98% crawling success rate for hard-to-scrape websites, significantly reducing the development effort required for data processing. The platform is extremely developer-friendly, offering native Python/JS SDKs and a 99.65% high-availability SLA to ensure the stability of production environments. It also supports unlimited concurrency by default with no extra fees for upgrades, catering to the high-throughput requirements of AI development. In addition, the platform features highly affordable pricing, making it an ideal choice for AI developers.

    1. 1

      Thank you so much for the detailed review! It means a lot to see these features being appreciated by a fellow engineer. My main goal is simply to keep the service stable and cost-effective so everyone can focus on building their agents without the extra headache. Glad it's helping with your workflow!

  18. 1

    Haha, honest confession: I'm actually one of the early seed users (been using it since June '25). The recent Reader API update has been a total game-changer for my workflows.

    To be fair, it was a bit rough in the beginning with quite a few bugs. But the founder was incredibly patient with my feedback and shipped fixes way faster than I expected. The guy clearly knows his tech! šŸ‘

    Huge potential here. Really hope it keeps staying indie-friendly for devs like us. Keep it up!

    1. 1

      Thank you so much! Honestly, with the pseudonym, I might not be able to guess exactly who you are, but I want to thank you from the bottom of my heart.Ā 

      Knowing that you stuck with me and the project during the most difficult early days means the world to me. It’s users like you who kept me going. Cheers to more updates ahead! šŸ„‚