Pricing & Cost Model

How access and price work on the hosted MCP. A fast-moving section with its own changelog. Exact per-gas numbers below stay placeholders; no payment flow is live yet.

Access cycle

The public server opens free at launch week. Payment is a later layer, and the line between free and paid moves with the hardfork calendar.

1. Open launch (5–9 October 2026). Hosted MCP, no x402, no API key. The free tier is the full Glamsterdam hardfork (EL alias Amsterdam) plus the Berlin→Fusaka lineage the lab already runs. Goal of the first weeks: real usage, and time to harden the open service.
2. Paid tier, a few weeks later. Once that open path has adoption data and has settled, we turn on x402 (USDC on Base). The delimiter is new EIPs that are not yet part of the hardfork on the free tier. The first paid capability we expect to ship is frame transactions (EIP-8141). Further interesting EIPs join this tier on a shorter, more automated cadence — that needs more of the EthereumJS fork pipeline than we have today.
3. Graduation. When the next hardfork (Hegota / Bogota) is on the horizon, the EIPs that were paid while they were ahead of mainnet — including EIP-8141 — move into the free tier as part of that hardfork. The next wave of post-fork EIPs starts paid again.

That loop is the product rhythm: today’s hardfork is open; tomorrow’s EIPs are paid until they become today’s hardfork.

Token holder discounts attach to the paid tier when it exists. They are never a gate, and they are not part of launch week.

Pricing model (draft, paid tier)

On the paid tier the direction is still linear pay-per-use via x402, in USDC on Base. The open hardfork stays free; we do not charge the Glamsterdam catalogue from request #1.

An earlier draft charged every request and skipped a free tier, on the theory that a sub-cent signature is the same friction as a free auth challenge. Launch week tests the other side of that: agents need a door they can walk through before we ask them to pay. Spam on the open tier is handled by hard ceilings, not by a quote. The paid tier can stay stateless — the quote is per call, with no usage account.

Price per simulated gas

We're leaning toward pricing by the EVM's native compute unit — gas — not per HTTP request. The caller would supply a gasLimit; the server would quote gasLimit × base_rate in the 402 response before touching the engine, and an out-of-gas halt would protect the worker.

• Base rate: TBD (e.g. ~$0.0000001 / gas — under discussion).
• Deep/multi-step endpoints could use an exponential curve so expensive queries self-price out of an agent's budget.

Discounts (token)

Token holders would get a discount on the gas price, not free access — see Token Utility. Indicative tiers (under discussion): $5 → 15%, $20 → 30%, $100 → 50%. Non-holders would simply pay the base rate; zero token friction.

Future: enterprise tier

A flat annual stablecoin subscription (e.g. ~$799 USDC/yr) for budget predictability and for firms that can't hold tokens — a hybrid SaaS model to introduce "when they come."

Anti-abuse (planned defenses)

Because agents are tireless cost-optimizers, defenses would be economic and architectural rather than human-friction based:

• Hard ceilings at the gateway reject queries beyond a max simulation depth — on the open tier and on the paid tier. Payment does not buy a larger ceiling.
• MCP schema guardrails instruct the agent not to issue oversized requests in the first place.
• On the paid tier, the quote makes spam economically self-limiting.
• Optional later: a minimum on-chain balance or ERC-8004 identity check if the open tier needs a Sybil brake beyond ceilings.

Cost model (early estimate)

Revenue would need to clearly exceed the cost of compute. Main drivers we're modeling:

• Compute — AWS EC2 for the headless MCP (the dominant variable cost; scales with simulation volume × gas).
• Hosting — the website remains on a low-cost Strato V-Server.
• Settlement — x402 facilitator fees.

The $100 discount tier is deliberately pitched near the real per-call cost level, so heavy users would be nudged toward holding the token while margins stay healthy. Detailed unit economics: to be modeled in a future round.

Changelog

Pricing Changelog

1. v0.62026-10-01Access cycle — free Glamsterdam at launch; x402 paid tier weeks later, starting with EIP-8141; graduation into the next hardfork.
2. v0.52026-09-02x402 decided for launch week (USDC on Base) — per-gas model unchanged; payment not live yet.
3. v0.42026-09-02Coupled to launch week — x402 target on public hosted MCP; engine exists, payment not live.
4. v0.32026-06-30Reframed as draft pricing model — no live payment flow; conditional language throughout.
5. v0.22026-06-30Decided: linear x402 per-gas pricing, no free tier; token = tiered discount; enterprise annual tier later.
6. v0.12026-06-30Initial scaffold — pricing/cost-model placeholders.

Add a one-line entry here whenever the pricing or cost model changes.