AWS & Hosting

Where the two legs run — and where they're heading for launch week. The headline: keep the website cheap; put the latency-sensitive MCP on dedicated compute.

Current state

Everything public today (the website, docs, community-token, roadmap, mcp-docs) runs behind nginx on a single Strato V-Server. That's fine for static sites and the explorations frontend.

The MCP engine and gateway are developed and tested locally and on AWS EC2 in preparation for the public HTTP endpoint — see server-config/aws/ for the bootstrap walkthrough. Public HTTP at mcp.feelyourprotocol.org is not launched yet.

Why the MCP needs dedicated compute

A V-Server uses shared (KVM) virtualization — fine for bursty web traffic, but a problem for sustained, CPU-bound EVM simulations. A "noisy neighbor" can spike CPU steal time, and an agent waiting on a 100ms simulation will time out (and may blacklist the tool) if it randomly takes seconds. Hosted agent traffic needs deterministic latency — which is why we're on a separate compute tier. x402 verification joins the main thread when the paid tier ships.

Target architecture (in progress for launch)

• Compute-optimized EC2, ARM/Graviton (c7g). Node.js + EthereumJS run very well on ARM; Graviton gives strong price/performance for heavy CPU work.
• Dedicated vCores mapped to the worker pool. Pin the Node worker_threads pool size to the instance's vCores so each isolated simulation gets predictable compute (see the API concept page).
• Main thread = traffic controller. Handles I/O and MCP routing; confirmed work goes to a worker that runs the bytecode and returns the trace. x402 verification lands here with the paid tier.

The hybrid setup

feelyourprotocol.org (website, docs, …)  →  Strato V-Server (nginx, low cost)
mcp.feelyourprotocol.org (hosted MCP)    →  AWS EC2 c7g (dedicated compute)  [launch week target]

Dedicated compute for the hosted lab, without over-engineering the educational pages.

Scaling boundaries

Scale horizontally for more concurrent isolated simulations (more workers / instances). What we explicitly do not plan to scale into is stateful, sequential multi-block historical processing — archive-node territory and outside our scope (see boundaries). Cost drivers feed directly into the cost model.

Open questions (later)

Region finalization, autoscaling policy, observability stack — to be decided after launch week stabilizes.

Changelog

AWS Changelog

1. v0.32026-10-01Dedicated compute is for the open hosted lab. x402 verification joins with the paid tier.
2. v0.22026-09-02EC2 bootstrap in progress — not nothing has moved; public HTTP still pending launch week.
3. v0.12026-06-30Initial hybrid Strato/AWS outline.