Roadmap

Condensed from the founding plan (notes/plans/roadmap-2026-07.md — the repository copy is canonical and more detailed).

Why this exists

Minecraft plugin/mod development accumulates automation piecemeal — a bot script here, a screenshot harness there, a playtest bootstrapper, CI jobs that boot real servers. This project unifies those layers, generalizes them beyond any one plugin, and puts them behind a single MCP tool surface an AI assistant can drive end to end.

Three lanes for driving a client

LaneWhat it isWhen
Agent modKotlin Fabric mod inside a real rendering client: widgets by name, keybinding invocation, movement, in-process screenshots, reflection stateThe main lane — live interactive testing
Protocol botMineflayer — a headless protocol-level playerCheap smoke and load work
OS levelWindow discovery, PrintWindow screenshots, open/closeAny client, vanilla included, nothing installed

Phases

PhaseDeliversStatus
0 — Scaffold & gatesMesh standards, working MCP skeleton, 3-layer test suite, coverage gate ≥90%, CI/CodeQL/Scorecard, this docs sitedone
1 — MCP core + OS layerJob model, instance registry, window discovery + OS screenshots, open/closenext
2 — Build/test orchestrationGradle driver, Paper provisioning/boot/console/logsplanned
3 — Control protocol + Paper agentOne loopback protocol spec; server-side state, commands, world ops, eventsplanned
4 — Fabric client agentSemantic UI driving by name, player control, in-process screenshots, opt-in gatingplanned
5 — Client lifecycleDev-client spawns (no launcher), attach handshake, Mineflayer laneplanned
6 — Deep state accessReflection gateway — reads session-gated, writes per-session grantplanned
7 — Release engineeringFull CI matrix incl. client gametests, GitHub-only releases with provenancepartially pulled forward (CI, Scorecard, this site)
8 — Extended targetsNeoForge agents, Folia-ready Paper agent, 26.x line when tooling catches upplanned

Deliberate constraints