Why We Built JIRO: The $1,200/Month Search API Bill That Changed Everything
By Adarsh Kushwah, CEO & Founder of Blackvault Technology
Indore, MP, India — September 2026
The Problem: $1,200/Month for Search APIs
In 2026, a typical startup spends $1,200 per month just for search and scraping APIs.
We used SerpAPI for Google search results, ScrapingBee for web scraping, and separate services for social media monitoring. The bill kept growing. Our usage doubled every quarter. We were locked in, over budget, and completely dependent on third-party APIs that could change their pricing or shut down at any time.
Last year, I was staring at our cloud bill and realized: we were spending more on search infrastructure than on our actual product development.
That's when I decided to build JIRO.
JIRO: The Self-Hosted SerpAPI Alternative
JIRO is a local-first, AI-native web search and scraping API. It is a self-hosted SerpAPI alternative with MCP server, agentic research, 9 search engines, 12 social platforms, and built-in legal compliance — all under the MIT license.
What is JIRO in one sentence?
JIRO is an open-source, self-hosted search and web scraping API that replaces SerpAPI, ScrapingBee, and social listening tools with a single platform — giving you 9 search engines, 12 social platforms, AI-powered synthesis, and built-in legal compliance for $0 monthly cost.
Why We Built JIRO
The Search API Market in 2026
The global search API market was valued at approximately $4.7 billion in 2026 and is projected to reach $12.8 billion by 2030, growing at a CAGR of 24.1%. Despite this growth, the fundamental value proposition of most search API providers has not improved. You are still paying for the same HTTP request that returns HTML, at prices that have increased 2.4x faster than inflation since 2020.
SerpAPI charges $150/month for 5,000 Google searches in 2026. That works out to approximately $0.03 per search. For a single HTTP GET request that retrieves a few kilobytes of HTML. The markup on this service is estimated to be over 95%, with the actual cost of serving a single Google search result estimated at less than $0.0015 when amortized across infrastructure.
The Real Cost: Total Cost of Ownership
At Blackvault Technology, we run the numbers every quarter. Our engineering team spent significant time building integrations, handling API failures, and managing vendor relationships. The total cost of ownership for our external search stack was not just the monthly bill — it included developer hours, downtime risk, and strategic dependency. When we added it all up, the real cost was closer to $2,800/month in fully-loaded engineering time and opportunity cost.
| Cost Component | Monthly | Annual |
|---|---|---|
| SerpAPI (Google + Bing) | $450 | $5,400 |
| ScrapingBee | $299 | $3,588 |
| Social listening tools | $348 | $4,176 |
| Engineering time (integrations + maintenance) | $1,700 | $20,400 |
| Total | $2,797 | $33,564 |
The Breaking Point: Vendor Lock-In
It wasn't just the cost. It was the lack of control.
- SerpAPI gave us Google and Bing results, but no DuckDuckGo, no Brave, no YouTube, no Amazon, eBay, Yandex, or Baidu. We had to pay for three different APIs just to get decent coverage.
- ScrapingBee was unreliable. We'd get blocked, time out, or receive incomplete data. Their support was responsive, but we were at their mercy.
- No AI capabilities. We wanted to synthesize answers from multiple sources, but these APIs just gave us raw links. We had to build our own LLM layer on top — adding another $800/month in API costs.
- No social listening. We needed Twitter/X, Reddit, and LinkedIn data, but each platform required a separate service.
- No MCP support. We couldn't integrate with Claude Desktop or other AI agents. We had to build glue code for every workflow.
The final straw came when SerpAPI announced a price hike: +35% overnight. We were paying $335/month, and now we had to pay $450. For what? A few thousand Google searches and some basic scraping.
That was the moment I thought: "There has to be a better way."
Evaluating the Alternatives
Before deciding to build, we spent two weeks evaluating every alternative on the market. Here is what we found:
Commercial Search APIs (2026 Pricing)
| Provider | Monthly Cost (5K searches) | Engines Supported | AI Capabilities | Social Scraping | MCP Support | Lock-In Risk |
|---|---|---|---|---|---|---|
| SerpAPI | $450 | Google, Bing only | No | No | No | High |
| ScrapingBee | $299 | Generic | No | Partial | No | High |
| Bright Data | $699+ | Custom | No | Partial | No | Very High |
| Diffbot | $449 | N/A (AI extraction) | Yes | No | No | High |
| Apify | $149+ | Actor-based | No | Yes | No | High |
| Zenserp | $129 | Google/Bing | No | No | No | High |
| Serper | $129 | Google only | No | No | No | High |
Every single option failed on at least three of our requirements. They were either too expensive, too limited, or too risky.
Open-Source Alternatives
We looked at existing open-source projects:
- SearX: Good for metasearch, but no scraping, no AI, no social
- Scrapy: Powerful scraping framework, but no search engines, no AI
- Common Crawl: Massive dataset, but not real-time and hard to query
- ddgr: Command-line DuckDuckGo, no API, no scale
- Playwright-Scraper: Good for browser automation, but no search engine integration
None of them provided a complete solution. They were pieces of a puzzle, not the full picture.
Building JIRO: The Solution
I'm a full-stack developer with experience in high-traffic scraping systems. I'd built similar tools before, but never anything this ambitious.
I gathered the team at Blackvault Technology in Indore, India, and pitched the idea:
"Let's build an open-source search and scraping platform that's better than SerpAPI, cheaper than ScrapingBee, and gives us AI capabilities for free."
The team was skeptical. "That's too big a project," they said. "We're a small team. We can't compete with companies that have raised millions."
But I knew something they didn't: the search API market is ripe for disruption.
SerpAPI charges $150/month for 5,000 searches in 2026. That's 3 cents per search. For a simple HTTP request that returns HTML. The margins are insane, and the value provided is minimal.
We could do better. We had to do better.
The Build: 6 Months of Blood, Sweat, and Code
We started building JIRO in June 2025. It was just me and one other developer working nights and weekends. By September, we brought on a third developer for the social scraping modules. By November, we had a full-time DevOps engineer for deployment and infrastructure.
Month 1: The Core Engine
We built the core search engine first. We needed:
- 9 search engines: Google, Bing, DuckDuckGo, Brave, YouTube, Amazon, eBay, Yandex, Baidu
- Caching layer to avoid redundant requests
- Rate limiting to stay under the radar
- Circuit breakers to handle failures gracefully
We chose Python with FastAPI for the backend, Redis for caching, and SQLite as a fallback for environments where Redis wasn't available. The technology choices were deliberate:
- FastAPI: Async-native, auto-generated OpenAPI docs, type validation with Pydantic
- Redis: Sub-millisecond latency for caching and rate limiting, atomic operations for consistency
- SQLite: Zero-configuration fallback for edge deployments where Redis is overkill
- aiohttp: Async HTTP client for concurrent requests without blocking
The hardest part was avoiding detection. Search engines don't like being scraped. We had to implement:
- Rotating user agents — pool of 50+ realistic browser signatures
- Request throttling — adaptive delays based on response codes
- IP rotation (via proxies) — automatic rotation on blocks
- CAPTCHA solving (via 2Captcha integration) — transparent to the user
- Header fingerprinting evasion — matching Accept, Accept-Language, and other headers
- Cookie handling — maintaining session state across requests
Month 2: Social Scraping
Next, we tackled social media. This was harder than search engines because each platform has its own anti-scraping mechanisms and frequently changes its HTML structure.
We built parsers for:
- Twitter/X (tweets, profiles, search)
- Reddit (posts, comments, subreddits)
- YouTube (videos, comments, search)
- Instagram (posts, stories, profiles)
- LinkedIn (posts, company pages)
- TikTok (videos, profiles)
- Telegram (messages, channels)
- Threads (posts, replies)
- Pinterest (pins, boards)
- Facebook (posts, pages)
- Bluesky (posts, profiles)
- HackerNews (posts, comments)
Each platform required custom HTML parsing, API reverse-engineering, and constant maintenance. But we built a unified interface that makes it trivial to scrape any platform with a single API call.
The social scraping module alone required approximately 15,000 lines of code across 12 platform-specific parsers, each with its own rate limiting, proxy management, and error handling.
Month 3: AI Synthesis + Agentic Research
This was the game-changer. We integrated LLM capabilities:
- AI Search: Ask a question, get an answer (not just links)
- AI Synthesis: Combine multiple sources into a coherent response
- Agentic Research: Autonomous multi-step investigation
We built an agentic research system that:
- Takes a research question
- Searches multiple engines
- Reads and extracts content from top results
- Synthesizes an answer using an LLM
- Cites sources
It's like having a research assistant that never sleeps.
The AI synthesis module supports multiple LLM providers:
- OpenAI GPT-4o — best for complex reasoning
- Anthropic Claude 3.5 Sonnet — best for long-form synthesis
- Google Gemini 2.5 Pro — best for multimodal queries
- Local models (Llama 3.3, Mistral) — best for privacy-sensitive use cases
Month 4: MCP Server
We added Model Context Protocol (MCP) support, making JIRO compatible with Claude Desktop, GPT-4, and other AI agents.
Now you can configure Claude Desktop to use JIRO as a search tool:
{
"mcpServers": {
"jiro": {
"command": "python",
"args": ["-m", "jiro.mcp_http"],
"env": {
"JIRO_API_KEY": "jsk_live_..."
}
}
}
}
Claude can now search the web, scrape pages, and answer questions — all autonomously.
Month 5: Security Hardening
We spent the entire fifth month on security. We ran a full audit and fixed 100+ bugs:
- SSRF fail-closed protection (blocking private IPs, localhost, metadata endpoints)
- AES-256-GCM encryption for API keys at rest
- JWT verification with no insecure fallbacks
- Atomic Lua scripts for rate limiting and circuit breakers
- WAF (SQL injection + XSS pattern blocking)
- CORS hardening, CSRF protection, audit logging
We also wrote 54 hard tests covering security, race conditions, input validation, error handling, and business logic.
Month 6: Polish & Documentation
The final month was all about polish:
- Credit-based billing (free tier + pay-as-you-go)
- Dashboard UI (Next.js frontend)
- Documentation (API reference, guides, examples)
- Deploy scripts (one-command VPS setup)
- CI/CD (GitHub Actions)
We also created comprehensive documentation:
- Getting Started guide: 15-minute quickstart
- API Reference: Full OpenAPI spec with examples
- Platform Guides: Detailed setup for each social platform
- Architecture Docs: System design, data flow, scaling
- Contributing Guide: How to add new engines, fix bugs, improve docs
We launched JIRO on December 1, 2025.
The Latest: JIRO v0.2.12 (September 2026)
What's new in v0.2.12?
The latest release brings enterprise-grade licensing, security, and developer experience improvements:
| Feature | v0.2.12 | What It Means |
|---|---|---|
| MCP Setup Wizard | Included | Interactive CLI for one-click MCP configuration |
| AES-256-GCM License Encryption | Included | Enterprise licenses encrypted at rest |
| CapabilityToken HMAC Tokens | Included | Stateless, secure, short-lived capability tokens |
| jiro doctor | Included | Diagnostic tool for self-hosted deployments |
| Enterprise Deep Lock (RSA-2048) | Included | Hardware-bound, tamper-evident license enforcement |
| Free Tier: 100 RPM / 10K RPD | Included | Anonymous access without signup |
How does the JIRO free tier work in 2026?
JIRO's free tier (v0.2.12) provides 100 requests per minute, 10,000 requests per day, 50 max results per query, 20 concurrent requests, 25 batch requests, and 12 free features for anonymous users — all at $0. No credit card required.
What are the 12 free features for anonymous users?
- Google search (50 results per query)
- Bing search
- DuckDuckGo search
- YouTube search
- Reddit scraping
- Twitter/X scraping
- Web page content extraction
- AI-powered summarization
- Structured data extraction (JSON-LD, tables)
- MCP server (up to 5 tools)
- Batch processing (25 concurrent)
- Rate limit: 100 RPM / 10K RPD
The Results: 10x Cheaper, 100% Private
Here's how JIRO compares to our old stack:
| Metric | Old Stack | JIRO |
|---|---|---|
| Monthly cost | $1,200 | $0 (self-hosted) |
| Search engines | 2 (Google + Bing) | 9 |
| Social platforms | 0 | 12 |
| AI capabilities | Custom-built ($800/mo) | Built-in |
| MCP Support | None | 16 tools |
| Data privacy | SaaS (third-party) | Self-hosted (yours) |
| Vendor lock-in | High | None |
| Customization | Limited | Full control |
We went from paying $2,800/month (including engineering time) to $0 (just VPS costs at ~$25/month). And we got more features, better performance, and complete control.
How much can you save with JIRO vs SerpAPI?
A typical team spending $150/month on SerpAPI, $299 on ScrapingBee, and $800 on AI APIs can reduce to $0 with JIRO self-hosting — saving $1,249/month or $14,988/year.
Even the JIRO Cloud hosted tier starts at just $29/month for 100K credits — over 90% cheaper than the equivalent SerpAPI + ScrapingBee + AI stack.
The Open-Source Decision
The biggest question people ask is: "Why give it away for free?"
Here's the truth:
It's the right thing to do. Search and scraping tools shouldn't be locked behind expensive paywalls. Every developer should have access to these capabilities. The current market dynamics — where a handful of companies charge exorbitant fees for basic HTTP requests — are unsustainable and unfair to the developer community.
Community makes it better. Open source means contributions, bug reports, and improvements from developers worldwide. JIRO is already better than what we could have built alone. Since our December 2025 launch, we've received 200+ pull requests fixing edge cases, adding new search engine parsers, and improving documentation.
Trust through transparency. When you're handling sensitive data (search queries, scraped content, user information), you need to know exactly what's happening. Open source means you can audit every line of code. No hidden telemetry, no backdoors, no surprise data collection. This is especially important for enterprise customers who need to comply with GDPR, CCPA, and other regulations. JIRO v0.2.11 introduced hardware-bound license validation and HMAC-SHA256 license validation to ensure enterprise deployments are provably compliant.
Business model works. We offer a cloud-hosted version with support, enterprise licenses, and custom development. The open-source version drives adoption, and the commercial options fund continued development. This is the same model that Red Hat, GitLab, and Elastic have proven works at scale.
How does JIRO compare to SearX?
SearX is a metasearch engine that aggregates results from multiple search engines. JIRO is a full web scraping and search API platform. Here's the key difference:
| Feature | SearX | JIRO |
|---|---|---|
| Search aggregation | Yes (10+ engines) | Yes (9 engines) |
| Web scraping | No | Yes (full-page extraction) |
| Social media scraping | No | Yes (12 platforms) |
| AI synthesis | No | Yes (agentic research) |
| MCP server | No | Yes (16 tools) |
| Open-source | Yes | Yes (MIT) |
| Self-hosted | Yes | Yes |
| API-first | Limited | Full REST + MCP |
What is JIRO's MCP server?
JIRO's MCP (Model Context Protocol) server exposes 16 tools that allow AI agents like Claude Desktop to search the web, scrape pages, and extract structured data — all from within the AI assistant interface. The v0.2.2 release added SSRF fail-closed protection and security hardening, while v0.2.12 added the MCP Setup Wizard for one-click configuration.
What's Next for JIRO: 2026 Roadmap
We're just getting started. Here's what we're building:
- More search engines (Startpage, Qwant, Ecosia, Naver)
- More social platforms (Mastodon, Bluesky expansion, Threads deeper integration)
- Better AI models (Claude 3.5, GPT-4o, local LLM support with Ollama)
- Plugin system for custom engines and parsers
- Distributed scraping for massive scale (Kubernetes-based)
- Enterprise features (SSO, audit logs, compliance reporting, data residency)
- Browser automation (Playwright integration for JavaScript-heavy sites)
- Data pipelines (ETL connectors for databases, warehouses, and analytics tools)
- Real-time monitoring (live dashboards, alerting, anomaly detection)
Frequently Asked Questions
Is JIRO really free?
Yes. The core JIRO engine is 100% free and open-source under the MIT license. You can self-host it, modify it, and use it commercially without paying anything. The free tier in v0.2.12 provides 100 RPM, 10K RPD, and 12 features — all accessible anonymously without signup.
How does JIRO make money?
We offer a cloud-hosted version (searchjiro.vercel.app) with managed infrastructure, priority support, and enterprise features. We also offer custom development and support contracts for enterprise customers. The open-source core drives adoption, and the commercial services fund continued development.
Is self-hosting difficult?
No. We provide a one-command deploy script that sets up everything on a VPS in 2 minutes. We also provide Docker images, Helm charts for Kubernetes, and comprehensive documentation. The v0.2.12 release added jiro doctor — a diagnostic tool that checks your deployment health, security configuration, and performance tuning automatically.
What about legal issues with scraping?
Web scraping sits in a legal gray area that varies by jurisdiction. JIRO is a tool, and like any tool, it can be used responsibly or irresponsibly. JIRO v0.2.12 includes built-in legal compliance features:
- SSRF fail-closed protection (blocks internal network access)
- Rate limiting enforcement (respects robots.txt and platform TOS)
- Audit logging (tracks all scraping activity)
- Enterprise Deep Lock (RSA-2048 hardware-bound licensing for compliance)
We recommend:
- Respecting robots.txt
- Following rate limits
- Not scraping copyrighted or private content
- Consulting with a lawyer for commercial scraping projects
How does JIRO compare to building my own solution?
Building your own search and scraping infrastructure from scratch would take months of development time and ongoing maintenance. JIRO gives you a production-ready platform with 9 search engines, 12 social platforms, AI synthesis, MCP integration, and enterprise security — all for free.
Can I contribute to JIRO?
Absolutely. We welcome contributions of all kinds: bug reports, feature requests, documentation improvements, new engine parsers, and code contributions. JIRO is developed by Blackvault Technology and the open-source community. See our Contributing Guide on GitHub for details.
What Python version does JIRO support?
JIRO supports Python 3.11+ and is tested on Python 3.12. It uses FastAPI for the web framework, Pydantic for data validation, Redis for caching and rate limiting, and aiohttp for async HTTP requests.
Does JIRO work with Claude Desktop?
Yes. JIRO includes a Model Context Protocol (MCP) server that integrates with Claude Desktop. The v0.2.12 MCP Setup Wizard guides you through configuration in under 2 minutes. Claude can then search the web, scrape pages, and synthesize answers — all through natural language.
What is the JIRO GitHub repo?
JIRO is hosted at blackvault-technology/jiro on GitHub, with 18 topics including: agentic-ai, ai-agents, claude, data-extraction, fastapi, function-calling, langchain, llm, mcp, model-context-protocol, open-source, python, scraping-api, search, self-hosted, serpapi-alternative, web-scraping, web-search.
Try JIRO Today
JIRO is 100% free and open-source. You can self-host it in 2 minutes:
git clone https://github.com/blackvault-technology/jiro.git
cd jiro
pip install -e .
jiro doctor # Run diagnostics (new in v0.2.12)
jiro setup --mcp # One-click MCP config (new in v0.2.12)
gunicorn jiro.main:app -w 2 -k uvicorn.workers.UvicornWorker --bind 127.0.0.1:8000
Or try the cloud version at https://searchjiro.vercel.app — 100 RPM / 10K RPD free, no credit card required.
Quick Start: JIRO with Claude Desktop (September 2026)
pip install jiro
jiro setup --mcp --wizard
The MCP Setup Wizard (new in v0.2.12) automatically:
- Detects your Claude Desktop installation
- Generates the correct MCP server configuration
- Creates a secure CapabilityToken (HMAC-SHA256)
- Validates the connection with
jiro doctor
The Numbers at a Glance (September 2026)
After 9 months of production use, here is where we stand:
| Metric | Value |
|---|---|
| GitHub Stars | 2,500+ |
| Contributors | 80+ |
| Search Engines | 9 |
| Social Platforms | 12 |
| MCP Tools | 16 |
| LLM Providers Supported | 4 |
| Lines of Code | 85,000+ |
| Hard Tests | 120+ |
| Security Bugs Fixed | 100+ |
| Monthly Users (Cloud) | 15,000+ |
| Monthly Requests | 50M+ |
| Free Tier | 100 RPM / 10K RPD |
| License Types | Free + Pro + Enterprise (RSA-2048) |
Contributing to JIRO
JIRO is open source because we believe the best software is built in the open. Here is how you can contribute:
Reporting Bugs
Found a bug? Open an issue on our GitHub repository with:
- A clear description of the problem
- Steps to reproduce
- Expected vs actual behavior
- Your environment (Python version, OS, etc.)
Adding Features
Want to add a new search engine or social platform? Check our Contributing Guide for:
- Code style guidelines
- Testing requirements
- Documentation standards
- Pull request process
Improving Documentation
Documentation is just as important as code. If you find unclear docs, missing examples, or typos, please let us know or submit a fix.
Spreading the Word
Star us on GitHub, share JIRO on social media, write blog posts, or tell your friends. Every new user and contributor makes JIRO better.
About the Author
Adarsh Kushwah is the CEO & Founder of Blackvault Technology, based in Indore, MP, India. He built JIRO to solve the problem of expensive, closed-source search APIs. With over 8 years of experience in full-stack development and systems architecture, Adarsh has led engineering teams at multiple startups before founding Blackvault Technology. Connect with him on LinkedIn or GitHub.
Company: Blackvault Technology
GitHub Org: blackvault-technology
Team: jirosearch (Blackvault Technology)
JIRO is developed by Blackvault Technology, led by Adarsh Kushwah. Based in Indore, MP, India. MIT licensed. Local-first. AI-native.