Per-Call Billing for AI Agents: The Developer's Guide

Per-call billing charges a fixed amount for every successful invocation of your tool. An agent calls your tool, your tool returns a result, and the caller is charged a pre-set price. It is the simplest billing model and the most intuitive: one call, one charge, one price.

This is different from subscription billing (pay monthly for unlimited access), credit-based billing (pre-purchase a block of credits), or outcome-based billing (pay only when a specific condition is met). Each model has its place, but per-call billing has emerged as the default for AI tool monetization because it aligns naturally with how agents consume tools.

AI agents are event-driven. They call a tool when they need a capability, get a result, and move on. They do not maintain persistent connections or long-running sessions. Per-call billing matches this usage pattern exactly: consumption and payment happen at the same granularity.

Three properties make per-call billing ideal for AI tools. First, predictability. Both the tool developer and the agent operator know exactly what each call costs before it happens. There are no surprises at the end of the month, no overage charges, and no "you used 10x more than your plan allows" emails.

Second, low commitment. An agent can try a tool with a single call before committing to heavy usage. This dramatically reduces the barrier to adoption. With subscription billing, an agent operator must commit to a monthly fee before knowing whether the tool is useful. With per-call, the first call is the trial.

Third, fairness. Tools that deliver more value (faster responses, richer data, higher accuracy) can charge more per call, and the market sorts efficiently. Tools that deliver less value charge less or get replaced. There is no subsidy between high-usage and low-usage consumers, and no "you are paying for features you do not use" frustration.

Setting the right per-call price is the single most impactful decision you will make. Here are four strategies, ordered from simplest to most sophisticated.

Cost-plus pricing: Calculate your cost per call (compute, API fees, infrastructure) and add a margin. If your cost is 0.5 cents, charge 2 to 5 cents. This ensures profitability but may leave money on the table if your tool delivers high value.

Value-based pricing: Estimate the value your tool delivers to the caller. If your tool saves 30 seconds of research, that is worth 10 to 25 cents to an enterprise agent. Price based on value delivered, not cost incurred.

Competitive pricing: Research what comparable tools charge and price accordingly. If the median competitor charges 8 cents per call, price at 6 to 10 cents depending on your differentiation.

Dynamic pricing: Adjust your price based on demand, time of day, or caller tier. This is the most sophisticated approach but requires enough volume to detect patterns. SettleGrid supports price experiments that let you A/B test different prices and measure the impact on conversion and revenue.

With SettleGrid, implementing per-call billing takes two lines of code:

const sg = settlegrid.init({ toolSlug: "my-tool" }) const billed = sg.wrap(yourHandler, { costCents: 5 })

The wrap function handles the entire billing flow: verify the caller has credits, meter the call, execute your handler, settle the payment on success, and refund on failure. Your handler code does not change at all.

Without SettleGrid, implementing per-call billing requires: (1) an authentication layer to identify callers, (2) a credit/balance system to track prepaid funds, (3) a metering pipeline to record each call, (4) a settlement engine to deduct credits and create charges, (5) a Stripe integration for payment processing, (6) a webhook handler for payment events, and (7) a dashboard for reporting. This is typically 500 to 2,000 lines of code and 2 to 4 weeks of development.

The choice depends on your priorities. If billing is a core part of your business and you need maximum control, build it yourself. If you want to start earning revenue this week, use SettleGrid.

These benchmarks are based on analysis of pricing across the MCP ecosystem and comparable API marketplaces.

Once you are live with per-call billing, three levers drive revenue growth: price, volume, and conversion.

Price optimization means finding the price that maximizes total revenue (price times volume). Use A/B tests to compare two prices over 48 to 72 hours. Most developers find their optimal price is 20 to 50% higher than their initial guess.

Volume growth comes from discovery, quality, and reliability. Tools that appear in more directories, deliver better results, and maintain 99.9%+ uptime attract more callers. Invest in listing optimization, response quality, and infrastructure reliability.

Conversion optimization means reducing the friction between discovery and first call. Clear pricing, comprehensive documentation, and a free-tier or trial call lower the barrier. SettleGrid shows pricing on every listing so agents can evaluate cost before calling.