In an industry where every API charges by the call — OpenAI, Twilio, Stripe, AWS, every SaaS API you've ever used — we did something that makes investors nervous and competitors confused: we gave away free API keys with unlimited call volume.
No rate limits. No per-call charges. No "free tier" that expires after 30 days. Just a key, a connection, and access to real-time hotel inventory from 2M+ properties, for as many calls as you can make.
People ask me the same question every week: "How can you afford to do that? Aren't you losing money?"
The short answer: no, we're not losing money. The longer answer is more interesting — it's a bet on a fundamentally different business model for AI infrastructure, one that turns the traditional API pricing playbook upside down. This is the inside story of why we did it, how it works, and what it means for the future of AI infrastructure pricing.
The Traditional API Pricing Model (and Why It's Broken for AI)
Let's start with how API pricing normally works. You build a useful API. You charge developers for each call — maybe $0.001 per call, maybe $0.01, maybe a monthly subscription with a call limit. Developers pay for what they use. You make money on volume.
This model works well for traditional APIs — SMS, payment processing, weather data, maps. The number of API calls is predictable, tied to business activity, and relatively stable. A company that sends 10,000 SMS messages a month knows it will send roughly 10,000 next month. Revenue is predictable.
But AI agents break this model. An AI agent doesn't make 10 API calls to book a hotel. It might make 50 — searching multiple destinations, comparing prices, checking availability, verifying details, processing payment, sending confirmations. And as agents get more autonomous, they make more calls, not fewer. An agent that plans a complete trip might make 500+ API calls across flights, hotels, restaurants, activities, and transportation.
If we charged per call, developers would be afraid to build agents that make lots of calls. They'd optimize for minimal API usage, which means less capable agents, which means worse products, which means less demand for our infrastructure. Per-call pricing creates a perverse incentive: it punishes exactly the behavior we want to encourage — agents that use our tools deeply and frequently.
This is the core insight that led to free unlimited keys: per-call pricing is a tax on agent sophistication. We want agents to be sophisticated. So we removed the tax.
The Business Model: Free Keys, Transaction Revenue
If we're not charging per call, how do we make money? The answer is transaction revenue.
Here's how it works:
Developers get a free API key. No cost, no limits, no credit card required. They connect their AI agent to our MCP server and start building.
Agents use our tools for search and comparison. This is free. An agent can search hotels, compare prices, check availability, read reviews — as much as it wants, at no cost. We don't charge for the "looking" part.
When a booking is completed, we earn a transaction fee. This is where we make money. Every time an agent books a hotel through our MCP server, we earn a small fee from the hotel or booking provider. It's the same model that OTAs use — they don't charge for search, they charge for booking.
Volume drives revenue. More agents using our infrastructure = more bookings = more transaction revenue. The free key is the customer acquisition channel. The transaction fee is the monetization.
This model aligns our incentives perfectly with developers. We want developers to build agents that make lots of calls, because more calls mean more searches, more comparisons, and ultimately more bookings. We don't penalize agents for using our tools — we reward them with better, deeper integration.
The Numbers: Why Free Keys Are Cheaper Than Paid Acquisition
Let me show you the math, because this is where the strategy becomes obvious.
Traditional API company:
- Customer acquisition cost (CAC): $200–$500 per developer (ads, content marketing, sales)
- Revenue per developer: $20–$100/month in API call fees
- Payback period: 6–12 months
- Risk: developers optimize for minimal calls to save money
Our model:
- Customer acquisition cost: ~$0 (free key, self-service, word-of-mouth)
- Revenue per developer: $0 until their agent makes a booking, then transaction fees
- Payback period: immediate (no acquisition cost to recover)
- Incentive: developers build agents that use our tools deeply, driving more bookings
The free key is not a "loss leader" that we'll eventually charge for. It's the customer acquisition channel. By making the key free and unlimited, we eliminate the friction that prevents developers from trying our infrastructure. A developer who might not have tried a paid API will try a free one. And once they've built their agent on our infrastructure, they're not going to rebuild it on a competitor's.
This is the same playbook that Stripe used (free to start, pay per transaction) and that AWS used (free tier, pay for usage). The difference is that we've removed even the usage-based pricing for non-transactional calls, because we want to maximize agent sophistication.
The Scale Effect: Why Marginal Cost Approaches Zero
You might be thinking: "But every API call costs you money in server infrastructure. Unlimited calls = unlimited costs."
That's true in theory. In practice, the marginal cost of an MCP search call is approaching zero, for three reasons:
Caching. Hotel search results don't change every second. We cache search results for short windows (2–5 minutes), which means repeated searches for the same criteria hit the cache, not the upstream inventory provider. Cache hit rates are 60–70% for common searches.
Aggregation. We aggregate inventory from multiple suppliers, which means we can spread the cost of upstream API calls across many downstream MCP calls. One upstream inventory query can serve dozens of MCP search requests.
Infrastructure efficiency. MCP is a lightweight protocol — search calls are small, fast, and stateless. The infrastructure cost per million search calls is in the single digits of dollars. At scale, the cost of serving free search calls is negligible compared to the transaction revenue they generate.
The key metric isn't "cost per call" — it's "cost per booking." If we serve 10,000 free search calls that result in 100 bookings, and each booking generates $5 in transaction revenue, that's $500 in revenue against maybe $2 in infrastructure costs. The free calls are the funnel. The bookings are the monetization.
What Competitors Don't Understand
When competitors see our free unlimited keys, they assume one of three things:
"They'll start charging soon." No. The free key is core to the business model, not a temporary promotion. We'll keep it free because it's the most efficient customer acquisition channel we have.
"They're losing money and will run out of funding." No. The transaction revenue model is profitable at scale, and we're growing transaction volume faster than infrastructure costs.
"Their service must be low quality." This is the most annoying assumption. Free doesn't mean low quality. Our MCP server provides real-time inventory from 2M+ properties, with confirmed pricing, real booking capability, and 99.9% uptime. The free key is a business model choice, not a quality signal.
What competitors don't understand is that in the AI infrastructure era, the winning strategy isn't to charge for access — it's to make access free and charge for transactions. The companies that win will be the ones whose infrastructure is so deeply embedded in agent workflows that every transaction flows through them. Per-call pricing creates friction. Free access creates dependency. Dependency creates transaction volume. Transaction volume creates revenue.
The Risks (and How We Manage Them)
I'd be dishonest if I didn't acknowledge the risks of this model:
Risk 1: Abuse. A bad actor could use a free key to make millions of calls, scraping our inventory or running up our infrastructure costs. We manage this through behavioral monitoring — unusual call patterns, non-human traffic signatures, and rate limiting at the account level (not the call level). The key is unlimited for legitimate use, but we reserve the right to throttle abusive usage.
Risk 2: Free riders. Some developers will use our free infrastructure for search and comparison, then route bookings through a competitor or direct to the hotel, avoiding our transaction fee. This is unavoidable in an open ecosystem. But the data shows that 80%+ of agents that search through our MCP server also book through it, because the booking flow is integrated and seamless. The friction of switching to a different booking channel is higher than the perceived savings.
Risk 3: Margin compression. As more MCP servers enter the market, transaction fees could be bid down. We manage this by focusing on quality — real-time inventory, confirmed pricing, reliable booking, customer support — rather than competing on price alone. The cheapest MCP server isn't the one that wins; the most reliable one is.
The Bigger Bet: AI Infrastructure Will Be Free
Our free unlimited keys are a bet on a bigger trend: in the AI era, infrastructure will be free, and monetization will happen at the transaction layer.
Think about it: Google doesn't charge for search. It makes money on the transactions that search enables (clicks, ad conversions, purchases). Stripe doesn't charge for its API. It makes money on the payments that flow through it. AWS has a free tier because it knows that once you're on the platform, you'll spend money on usage.
AI infrastructure is following the same path. The MCP servers, the tool interfaces, the search and comparison capabilities — these will all be free. The money will be made on the transactions: bookings, payments, purchases, reservations. The companies that own the transaction layer will be the profitable ones. The companies that charge for access will be left behind.
This is why we offer free unlimited keys. It's not a promotion. It's not a mistake. It's a deliberate bet that the future of AI infrastructure is free access + transaction revenue, and that the companies that establish themselves as the free infrastructure layer today will own the transaction revenue tomorrow.
The key is free. The calls are free. The bookings are where we make money. And as more agents connect, more bookings flow through our infrastructure, and the flywheel spins faster.
That's the business model. That's why we dare to offer unlimited calls. And that's why the competitors charging per call are going to look increasingly outdated.
If you want to see the free MCP infrastructure model in action — and build an agent that can search, compare, and book hotels at no cost — the quick-start guide gets you connected in five minutes: RollingGo MCP Quick Start.
Top comments (0)