Problem & Vision

About this site

roadmap.feelyourprotocol.org is the Phase 3 strategy and history workspace — vision, tracks, timeline, and draft concepts for the sustainable business layer. It records where our thinking stands, including open questions, so we have a stable place to iterate.

What exists today: the explorations website, website docs, MCP docs, the execution engine and gateway (tools implemented), and years of EthereumJS work behind both legs. What we're shipping next: the public hosted MCP at mcp.feelyourprotocol.org, open and unpaid, with the full Glamsterdam hardfork — launch week 5–9 October 2026. x402 for newer EIPs follows after that service has usage and has been hardened.

This scope is deliberate: we are not building general mainnet infrastructure, RPC, or testnet ops — that work lives elsewhere in the ecosystem. FYP targets the gap between probabilistic LLMs and deterministic simulation of protocol changes not yet on mainnet.

For tracks, horizons, and how ideas evolve over time, see Roadmap & Tracks — fast-moving sections there and on the Agent API concept and pricing model pages each carry a micro-changelog at the bottom.

The Problem

There is a large and growing gap between the Ethereum protocol layer and the application layer. Protocol changes — EIPs, hard forks, and research — are hard to follow, hard to explain, and hard to build against. Feel Your Protocol started to help close that gap: to explain and educate on protocol updates, hands on.

Phase 3 adds a second axis to the same problem. As AI agents take on real protocol work — auditing contracts, optimizing MEV, testing upgrades — they hit a hard wall:

LLMs are probabilistic; the Ethereum protocol is strictly deterministic.

An AI model cannot reliably simulate cascading state changes, calculate exact gas under an unreleased EIP, or trace deep stack manipulations. It will confidently hallucinate. What agents lack is a deterministic ground-truth oracle they can call to get the real answer.

The Vision

Build Feel Your Protocol into the deterministic oracle for the future Ethereum protocol — a headless MCP server that wraps the modular EthereumJS stack so that AI agents (and the researchers behind them) can run real, exact simulations of upcoming fork rules, EIPs, and research targets on demand.

In one line: deterministic truth for probabilistic machines.

The lab equipment exists — engine, gateway tools, and a growing EIP catalogue on mcp-docs. The product milestone is the hosted endpoint agents connect to without self-hosting — open at launch, with a paid tier for EIPs that are still ahead of the current hardfork. Concrete tool schemas and limits live on mcp-docs; this site keeps the strategic sketch.

The educational website does not get left behind — it remains the visual front door and DevRel engine while keeping its teaching mission. See Two Legs, One Engine.

Humans (researchers, educators, integrators, the token community) and agents themselves are equal customers. Agents will discover and pay for services without a person who has already allowlisted us. Payment, onboarding, and discovery are built for that — see Two Audiences.

Why us (the moat)

• Domain depth. Years of building EIP prototypes and maintaining critical infrastructure inside the Ethereum ecosystem — context an LLM cannot synthesize on its own.
• A uniquely suited stack. EthereumJS is highly modular (13+ libraries), TypeScript-native, and exceptionally observable — easy to switch fork contexts, manipulate state, dump step-by-step traces, and expose cryptographic primitives (via the Noble ecosystem). This is hard to replicate with a monolithic Rust node.
• Founder & brand trust. A recognizable, education-first brand and an open-source track record lower the barrier for developers to trust — and allowlist — the infrastructure.
• Proof over promise. We document without MCP vs with MCP comparisons — same prompt, same model, different outcome — so the oracle thesis is checked, not asserted.

Target product — what we're aiming for (and what we're not)

Boundaries we hold to keep scope honest:

• Aiming for: a headless, stateless EVM simulation & cryptographic oracle. Bring-your-own-state; raw bytecode in, deterministic JSON trace out. Delivered as a hosted MCP server (not a self-host tutorial).
• Explicitly not: a generic RPC provider, an archive node, or a high-throughput indexer. We deliberately avoid sequential multi-block historical processing (see Tech Readiness & Boundaries).

These lines will move as we learn — operational detail lives on mcp-docs.

Where this fits

• Two Legs, One Engine — how the website and the MCP server reinforce each other.
• Two Audiences — humans and agents as equal customers.
• Principles & Operating Discipline — the guardrails we hold ourselves to.
• Roadmap & Tracks · Timeline · Launch week — how and when we ship.

Changelog

Problem & Vision Changelog

1. v0.62026-10-01Next ship is the open Glamsterdam MCP. Paid x402 tier follows for EIPs ahead of that hardfork.
2. v0.52026-09-24Humans and agents as equal customers; payment, onboarding, and discovery follow from that.
3. v0.42026-09-02Lab equipment built — public hosted launch is the next milestone; mcp-docs and generic MCP tools acknowledged.
4. v0.32026-07-15MCP docs site live — strategic sketch stays here; concrete docs on mcp-docs.
5. v0.22026-06-30Reframed as conceptualization workspace — conditional language for unshipped API.
6. v0.12026-06-30Initial problem & vision outline.