Skip to main content
Glama

OmniMod MCP — Professional Map & Mod Authoring for Any AI Agent

Version Tools Resources Node License Free

An MCP server that connects any AI agent (Claude Desktop / Claude Code / Cursor / Cline / Kilo Code / Zed / Windsurf / custom clients) to a live OmniMod game, gives it expert-level knowledge of the 1.8.8 engine and the Forge 1.20.1 mod-compat layer, and turns it into a professional map builder and mod author.

┌──────────────────┐     MCP (stdio JSON-RPC)    ┌──────────────────────┐
│ Any AI agent     │ ───────────────────────────▶│ OmniMod MCP server   │
│ (Claude, Cursor, │   tools / resources /       │   • 1.20→1.8 names   │
│  Cline, Kilo,    │   prompts                   │   • shape generators │
│  custom ...)     │ ◀───────────────────────────│   • batch builder    │
└──────────────────┘                             │   • mod scaffolder   │
                                                  │   • JAR inspector    │
                                                  │   • live game HTTP   │
                                                  │      :26911 bridge   │
                                                  └──────────┬───────────┘
                                                             │ HTTP
                                                  ┌──────────▼───────────┐
                                                  │ OmniMod game running │
                                                  │ (Desktop / Android / │
                                                  │  Web) — Agent Link   │
                                                  └──────────────────────┘

Why this exists

OmniMod runs on an EaglercraftX 1.8.8 engine but accepts Forge 1.20.1 mods. That sounds simple. It is not. The engine registers block / item names from 1.8 (minecraft:stone, minecraft:planks, minecraft:wool + meta 14 = red). Modern mod authors think in 1.20 names (minecraft:oak_planks, minecraft:white_wool, minecraft:grass_block).

There is no alias table on the placement path in the engine. Send minecraft:oak_planks and the op logs unknown_block and silently builds nothing. The batch is still recorded as applied. You will not see an error.

This server is the layer that closes that gap. It also gives the agent:

  • Every high-level map primitive a real builder wants (cylinders, spheres, gable and hip roofs, gabled houses with doors and window bands, schematic blueprints) — each compiled into the smallest possible set of engine-friendly fill_area / place_block ops.

  • A static knowledge base of every 1.8.8 block, every 1.8.8 item, the 1.8-era command surface, recipe JSON keys, mod-scaffolding conventions, and the most common silent-failure pitfalls — every fact cited with the engine file:line that proves it.

  • A JAR / folder inspector that lints mods against the loader's actual code and reports problems before they reach the device.

  • A mod scaffolder that writes a complete Forge-shaped folder tree from a JSON spec, using the exact keys the loader reads.

Related MCP server: aetherforge

Install with ONE command (Windows, free)

Open PowerShell and paste this single command — it downloads the MCP to the correct path (%LOCALAPPDATA%\OmniModMCP\omnimod-mcp), builds it, verifies it, and registers a global omnimod-mcp command that works from any folder in any terminal:

powershell -NoProfile -ExecutionPolicy Bypass -c "$f=$env:TEMP+'\omnimod-install.ps1'; irm https://raw.githubusercontent.com/Mcamento8/omnimod-mcp/main/scripts/install-windows.ps1 -OutFile $f; & $f"

(paste as ONE line in cmd or PowerShell — downloads the installer to a temp file and runs it; re-run any time to update)

Then, from any terminal, anywhere:

omnimod-mcp          # prints YOUR personal connect guide (Kilo / Cline / Cursor / Claude)
omnimod-mcp doctor   # checks node, files, game bridge, token, repo links

Re-run the same one-command any time to update to the latest version.

Alternative installs (no account, no cost):

Windows: double-click install.bat. macOS/Linux: run ./install.sh.

Manual install:

cd mcp
npm install
npm run build
npm run selfcheck   # runs the in-process test battery (registry, ops, shapes, scaffold, inspect, mapdocs)
npm run e2e        # end-to-end JSON-RPC test (68 tools, 13 resources, 5 prompts)

The server speaks MCP over stdio. Point your client at node dist/index.js. Ready-made configs for every major client live in examples/ (Claude Desktop, Claude Code, Cursor, Cline, Kilo/ZCode — plus npx_mcp.json for the npm route). The engine-source mirrors are pre-linked by default, so omni_knowledge {topic:"repos"} works with zero setup:

Publishing: see PUBLISHING.md for the step-by-step guide (publish the two engine-source repositories in repos/, link them via env vars, then publish the MCP itself — with the full verification battery).

The professional map-dev master guide: docs/MAP_DEV_MASTER_GUIDE.md — served to agents as the omni_map_guide tool and the omnimod://knowledge/map-dev-guide resource.

Running it: the three commands you actually use

omnimod-mcp setup     # print the CONNECTION CARD: every value you need, plus
                      # paste-ready JSON for Claude Desktop, Claude Code, Cursor,
                      # VS Code, Cline, Kilo Code, Windsurf, Zed, Continue, and a
                      # plain stdio block for anything else.
omnimod-mcp doctor    # diagnose a connection problem (node, files, bridge, token,
                      # repo links, cache folders)
omnimod-mcp serve     # RUN the server in THIS window and stay open.
                      # Close the window (or Ctrl+C) to stop it.

Normal use does not need serve. Your MCP client spawns the server itself — you paste the client block once, restart the editor, and that is it. No window to keep open.

serve exists for when you want to watch it: it prints the connection card and then runs the server in the foreground, so you can see the startup state (bridge URL, token, project/worlds/work paths, which asset libraries are linked) and watch it stop cleanly when you close the window. Do not type into that window — its input is the server's JSON-RPC channel.

The Windows launcher (install.bat) and the one-command installer both end by printing the card and offering to start the server for you; on macOS/Linux, ./install.sh --serve does the same non-interactively.

Configuring your client

The server needs three env vars (all optional — they have sensible defaults):

Env var

Default

Meaning

OMNIMOD_HOST

127.0.0.1

The IP / hostname of the device running OmniMod

OMNIMOD_PORT

26911

Agent Dev Link port (the engine default)

OMNIMOD_TOKEN

(none)

32-hex pairing token. If unset, call omni_pair to obtain one.

OMNIMOD_TIMEOUT_MS

20000

Per-request HTTP timeout. The engine bounds game-thread work at 10s.

OMNIMOD_PROJECT_ROOT

(none)

Absolute path to the OmniMod checkout, for local JAR/docs access.

OMNIMOD_WORK_DIR

~/.omnimod-mcp

Where the scaffolder writes mod folders by default.

OMNIMOD_AUTO_TRANSLATE_BLOCKS

true

Translate 1.20→1.8 names on the way out.

OMNIMOD_FORGE_COMPAT_REPO

(none)

Public GitHub mirror of the engine's Forge 1.20.1 compat layer (see omni_knowledge topic='repos').

OMNIMOD_COMMAND_BLOCKS_REPO

(none)

Public GitHub mirror of the engine's command-block / command-surface system (see omni_knowledge topic='repos').

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "omnimod": {
      "command": "node",
      "args": ["C:/path/to/OmniMod/mcp/dist/index.js"],
      "env": { "OMNIMOD_HOST": "192.168.1.42" }
    }
  }
}

Claude Code (.mcp.json in your project root or ~/.claude)

{
  "mcpServers": {
    "omnimod": {
      "command": "node",
      "args": ["C:/path/to/OmniMod/mcp/dist/index.js"]
    }
  }
}

Cursor (Settings → MCP → Add new global MCP server)

{
  "omnimod": {
    "command": "node",
    "args": ["C:/path/to/OmniMod/mcp/dist/index.js"]
  }
}

Cline (VS Code settings.json, "cline.mcpServers")

{
  "cline.mcpServers": {
    "omnimod": {
      "command": "node",
      "args": ["C:/path/to/OmniMod/mcp/dist/index.js"],
      "disabled": false
    }
  }
}

Kilo Code (ZCode) — ~/.zcode/config.json

Kilo Code is the original OmniMod project harness. Add the same shape to mcpServers:

{
  "mcpServers": {
    "omnimod": {
      "command": "node",
      "args": ["C:/path/to/OmniMod/mcp/dist/index.js"]
    }
  }
}

First agent call

Once the server is connected, the agent has 68 tools (plus the full map-dev master guide: omni_map_guide / resource omnimod://knowledge/map-dev-guide). A typical first session:

1. omni_map_connect {map: "my_map"}   -> ONE CALL: reads _dev/state/agentlink.json out of the
                                          map folder and sets host/port/token by itself
                                          (skip 1-2 below; no pairing code needed)
2. omni_ping                          -> confirm the bridge is up (pre-auth)
3. omni_pair {code: "ACDM3491"}       -> ONLY if you have no folder access: user reads the
                                          8-char code off the device
4. omni_state                         -> confirm a world is loaded (or quit+create)
4. omni_knowledge {topic: "rules"}    -> read the non-negotiables
5. omni_map_docs {map: "my_map"}      -> fetch the per-map context pack
6. omni_blueprint { apply: true,
                    steps: [
                      { kind: "cylinder", block: "minecraft:stone",
                        center: [0,64,0], radius: 12, height: 1 },
                      { kind: "building", origin: [0,65,0], width: 9, depth: 7,
                        height: 4, wallBlock: "minecraft:cobblestone",
                        roofStyle: "gable", roofBlock: "minecraft:stone" }
                    ]}
7. omni_logs {level: "WARN"}         -> read what the engine applied
8. omni_agentlog {message: "phase 1 done"}

The simple way to connect: the map folder IS the credential

Every OmniMod dev map carries a game-owned file:

worlds/<map>/_dev/state/agentlink.json   { mapName, port, token, scope:"map", enabled, endpoints }

Anything that can read that folder already holds full control of that map, so the handshake is gone: present the token as Authorization: Bearer <token> (or X-Agent-Token) and you are in — from any IP, with no pairing window, no 8-character code, no expiry and no re-pairing after a restart. omni_map_connect does it in one call; without the MCP it is one curl.

  • Scope: the credential is bound to its map. /omni/world/create and /omni/world/enter <other map> answer scope_violation; everything inside the map — build, commands, walking, entities, screens, and POST /omni/world/restart — is allowed.

  • Persistence: the token lives in the map folder, so it survives world restarts, game restarts and reboots. The engine re-binds the listener automatically while a dev map is loaded, so the connection self-heals instead of dropping.

  • The only off switch is the player: Options → Map Folder Agent Access. Turning it off answers bad_token to folder-based agents (batches still apply — only live control stops); turning it back on restores exactly the same access.

Tool catalog

Connection

  • omni_map_connect — [2026-09-16] the one-call connection: read the game-owned worlds/<map>/_dev/state/agentlink.json and point the session at that map's bridge (host + port + token). Whoever can read the map folder already holds the credential, so no pairing code is involved. Finishes with a live ping + state.

  • omni_ping — liveness probe (no auth)

  • omni_pair — quick-pair with an 8-char code (no auth)

  • omni_config — view or update host/port/token/autoTranslateBlocks

  • omni_state — full snapshot of game state

  • omni_help — endpoint catalog the device actually serves

  • omni_logs, omni_errors, omni_notifications — observability

  • omni_agentlog — write your own annotation into the log ring

  • omni_devpatch_verify — compare source sha256 against the active DevPatch

Worlds

  • omni_worlds, omni_world_create, omni_world_enter, omni_world_quit

  • omni_world_restart — [2026-09-16] reload the running map (quit → re-enter the same world, wait until playable). Applies pending _dev/ batches + re-runs the map's load function; use it for staged mods, <ns>:load wiring and anything that needs a fresh load.

  • omni_mapdev_status — apply-ledger state

Map building

  • omni_block_translate, omni_item_translate, omni_block_search

  • omni_shape_solid_box, omni_shape_hollow_box, omni_shape_cylinder, omni_shape_sphere, omni_shape_pyramid, omni_shape_gable_roof, omni_shape_hip_roof, omni_shape_building, omni_shape_line

  • omni_blueprint — compose many shape calls in one round-trip

  • omni_batch_validate, omni_batch_apply

Player & observation

  • omni_command, omni_player (teleport, look, give, attack, use, ...), omni_inventory, omni_world_scan, omni_world_raycast, omni_chat

Map development & per-map context

  • omni_map_onboard — show all agent docs and mandatory setup for a map

  • omni_map_docs — fetch the per-map context pack (knowledge + agent guidelines)

  • omni_map_bootstrap — seed a map folder with agent knowledge pack

  • omni_map_status — check map state, apply ledger, living doc status, and verdicts

  • omni_map_filemap — regenerate FILE_MAP.md from the build folder

  • omni_map_changelog — append to the append-only CHANGE_LOG.md or read it back

  • omni_map_overview — patch a numbered section in MAP_OVERVIEW.md

  • omni_map_report — write a verification report with a normalized path

Mod authoring

  • omni_mod_add — stage a JAR into a world

  • omni_mod_scaffold — write a Forge-shaped mod folder from a spec

  • omni_mod_inspect — lint a JAR or folder

  • omni_recipe_validate — static recipe linter

Knowledge

  • omni_knowledge — programmatic access to the static knowledge base

  • 11 MCP resources (one per knowledge topic + 1.8.8 block/item lists + per-map context pack)

  • 5 MCP prompts (build a medieval village, work on a map, verify a map change, author a new mod, debug a silent no-op)

Honest boundaries

  • The bridge does NOT run on Web targets. The _dev folder bridge is the alternative on Web (docs/project_map/30_AGENT_DEV_BRIDGE_PIPELINE.md). Use the desktop / Android target for the MCP.

  • The bridge does NOT encrypt. Use a trusted LAN or VPN.

  • The bridge does NOT change game logic. It calls into the existing command manager, the existing player APIs, the existing mod-staging path. If the game does not understand a command you sent, you'll get command_failed back.

  • The bridge does NOT run faster than the server. Every command runs on the integrated server thread with a 10-second latch. A heavy command will time out at server_thread_timeout.

  • The bridge does NOT manage per-map agent context. Use the omni_map_* tools to bootstrap a map folder with the agent knowledge pack and track living documents (MAP_OVERVIEW.md, CHANGE_LOG.md, FILE_MAP.md, verification reports).

Source layout

mcp/
├── src/
│   ├── index.ts          # entry point (start or selfcheck)
│   ├── selfcheck.ts      # smoke-test battery (registry, ops, shapes, scaffold, inspect, mapdocs)
│   ├── server.ts         # 51 MCP tools, 13 resources, 5 prompts
│   ├── bridge.ts         # HTTP client with error mapping
│   ├── config.ts         # env-driven runtime config
│   ├── translate.ts      # 1.20→1.8 block/item name+meta translation
│   ├── registry.ts       # 198 blocks + 187 items registered in 1.8.8
│   ├── ops.ts            # MapDev op validation + batch builder
│   ├── shapes.ts         # solidBox, cylinder, sphere, building, etc.
│   ├── scaffold.ts       # mod folder scaffolder
│   ├── inspect.ts        # mod JAR/folder inspector + problem linter
│   ├── knowledge.ts      # static facts (rules, pitfalls, commands, recipes, …)
│   └── mapdocs.ts        # per-map agent context pack and dev workspace tools
├── package.json
├── tsconfig.json
└── README.md

License

MIT — see LICENSE. Free for everyone, including commercial use. The translation table was extracted from the engine's own Blocks.java / Items.java registration calls, not from any third-party dataset. The engine-source mirrors linked by this server carry their own mirror license — see omnimod-forge-compat and omnimod-command-blocks.

Available Tools

51 tools
omni_agentlogA

Write your own annotation into the log ring. Use this to mark phases of an agent run (e.g. phase 1: scaffolding world).

Inputs: message (required), level.

omni_batch_applyB

Translate + validate + send a raw MapDev op batch to the bridge. The engine applies it within ~2s of the world being loaded.

Inputs: ops (required), name, dryRun, filename, +1 more.

omni_batch_validateA

Translate and validate a raw MapDev op batch WITHOUT sending it. Returns the cleaned batch + a list of errors and warnings (e.g. unknown block names, region over 1M blocks, y out of range, op count over 20k).

Inputs: ops (required), name, autoTranslate.

omni_block_translateA

Translate ONE 1.20 block id to its 1.8 form (with metadata). Returns translated, passthrough, or unresolved. This is the function to call before any 1.20 block name goes into an op — there is NO alias table on the placement path in the engine.

Inputs: id (required), meta.

omni_blueprintA

Run any sequence of high-level shape calls in a single MCP round-trip. Provide a list of named operations; the response is the full op list ready to feed into omni_batch_apply (or auto-applied if apply=true).

Inputs: steps (required), apply, batchName.

omni_chatB

Broadcast a blue [Agent] chat line.

Inputs: message (required).

omni_commandA

Run any in-game command (full-privilege, player-anchored). Response includes chat feedback.

Inputs: command (required).

omni_configA

View or update the live MCP connection config. host/port/token changes affect every subsequent bridge call in this session. Use this to point at a different device, or to set a token obtained by omni_pair.

Inputs: host, port, show, token, +3 more.

omni_devpatch_verifyA

Compare source-file sha256 against the active DevPatch inventory on the device. Returns mismatches (changed_since_patch), missing (tracked but not sent), and matches. NO-OP envelope when no devpatch is active.

Inputs: files.

omni_errorsB

Convenience for /omni/logs with min severity hardcoded to WARN. Use this first when something looks wrong.

Inputs: q, limit, since, source.

omni_helpA

Self-describing endpoint catalog the running device serves (the engine returns its own list). Use this to discover platform-specific additions.

Inputs: no named parameters.

omni_inventoryA

Full inventory, hotbar, and armor (slot id+label+item id+count+meta+damage).

Inputs: no named parameters.

omni_item_translateA

Translate ONE 1.20 item id to its 1.8 form (with metadata). Same rules as omni_block_translate but for the item registry (187 entries).

Inputs: id (required), meta.

omni_knowledgeB

Fetch the static OmniMod knowledge base. Topics: identity, rules, pitfalls, commands, recipe shapes, staging paths, endpoint catalog, troubleshooting.

Inputs: topic (required), pitfallId.

omni_logsB

In-game log ring (TRACE..FATAL + AGENT). Supports since (sequence cursor), limit, level (min severity), source substring, q substring, tail (file tail bytes).

Inputs: q, tail, level, limit, +2 more.

omni_map_bootstrapA

Create or repair the agent context pack inside a map folder. Static knowledge docs are (re)written when missing or version-stale; MAP_OVERVIEW.md / CHANGE_LOG.md / FILE_MAP.md are seeded only when absent and never overwritten. Use this for map folders the running game has not provisioned (copies pulled off a device, exported maps, Web-target saves).

Inputs: map, force, mapDir, worldsDir.

omni_map_changelogA

Append an entry to a map's CHANGE_LOG.md (append-only: existing history is never rewritten). Required at the end of every task. To correct an earlier entry, set corrects and append a new entry instead of editing the old one. Pass read: true to read the log instead of writing.

Inputs: map, read, agent, notes, +11 more.

omni_mapdev_modeA

Dual-mode switch for the running map (the DEV <-> PLAY preview system). Call with no args to QUERY the current mode. mode='play' (alias 'preview') = PREVIEW: every command block becomes invisible, unopenable, unbreakable and untargetable while ALL command-block logic (impulse/repeating/chain, redstone, functions, schedulers) keeps running — the map behaves exactly as published. mode='dev' = full editing shape. Switching is instant (chunk meshes rebuild), race-free (server-thread scheduled) and persisted per map in mapmode.json. After switching to PLAY, always verify the map still WORKS (hiding never disables logic). Same unified kernel as /omni_dev mode and the pause-menu Map Dev screen.

Inputs: by, mode.

omni_mapdev_statusB

MapDev apply-ledger state and current map name.

Inputs: no named parameters.

omni_map_docsA

Read specific documents from a map's agent context pack. Use omni_map_onboard first; use this for targeted lookups afterwards (e.g. just the troubleshooting file, or just the change log).

Inputs: map, files, mapDir, worldsDir.

omni_map_filemapA

Regenerate FILE_MAP.md for a map from the real folder contents: every file under _dev/ with owner/size/mtime, the world data files marked off-limits, every batch with its label and op count (flagging unparseable ones), the verification reports, and the sibling mod-staging folders. Run this after creating, renaming or deleting ANY file in the map.

Inputs: map, agent, mapDir, worldsDir.

omni_map_guideA

The complete professional map-development master guide (same content as the omnimod://knowledge/map-dev-guide resource): the 8-phase workflow, the full command surface (modern MCBP syntax + command-block modes), map functions, dual-mode DEV/PLAY testing, verification battery and handoff protocol. Read this once per session before building a map; use omni_knowledge(topic='mapdev') for the compact phase summary.

Inputs: no named parameters.

omni_map_onboardA

START HERE for any map task. Locates the map folder, verifies the per-map agent context pack, and returns the mandatory documents (mandate + map overview + change log + file map + build system + block names + testing mandate) as text so you actually read them before building. Also reports missing/outdated/unwritten documents. Run this before your first batch on a map, even when connected to the bridge — the MCP knowledge base describes the engine, this pack describes THIS map.

Inputs: map, full, mapDir, worldsDir, +1 more.

omni_map_overviewA

Read MAP_OVERVIEW.md, or replace one of its numbered sections. This is where the user's actual goal, the build specification, and this map's coordinate conventions live — write it BEFORE building, and correct it whenever a previous agent recorded the goal wrongly. Sections: 1 Identity, 2 The user's goal, 3 Current build specification, 4 Coordinate conventions, 5 Mods in this map, 6 Open items, 7 Do not touch, 8 Notes.

Inputs: map, body, mapDir, section, +1 more.

omni_map_reportA

Write a verification report into worlds//_dev/state/verification/. Every task owes one. Pass template: true to get the blank report skeleton to fill in.

Inputs: map, slug, mapDir, content, +2 more.

omni_map_statusA

Verify a map's context pack and workspace without changing anything: which documents exist, which are version-stale, which living documents are still unwritten, what batches exist, what the apply ledger says (ops/placed/failed per batch — Level 1 verification without the HTTP bridge), and which verification reports have been written.

Inputs: map, mapDir, worldsDir.

omni_mod_addA

Stage a mod JAR (base64) into a world. The next world load picks it up. If a world is running, reload it via omni_world_quit + omni_world_enter to activate.

Inputs: dataB64 (required), filename (required), world.

omni_mod_inspectB

Inspect a Forge mod (JAR or unzipped folder). Returns metadata, file counts, parsed models/recipes/GUIs, and a problem list (each problem cites the engine code that would trip on it).

Inputs: path (required).

omni_mod_scaffoldA

Write a Forge 1.20.1-shaped mod folder to disk. Output is a folder (not a ZIP) — zip it yourself before omni_mod_add, or stage it via the folder mod path on the engine side. Every JSON key written is one the engine actually reads (see scaffold.ts comments).

Inputs: spec (required), outDir (required).

omni_notificationsB

Grouped WARN/ERROR/FATAL by source. Compact issue feed (one row per source with count + sample).

Inputs: since.

omni_pairA

Quick-pair with the device using the 8-character code the user sees on Options -> Agent Link Info -> Quick Pair Computer. NO AUTH. After this call, the token returned is stored in the MCP's live config so the next call works without re-pairing.

Inputs: code (required).

omni_pingA

Liveness probe for the OmniMod Agent Dev Link bridge. NO AUTH. Returns service info, protocol version, and auth requirement.

Inputs: no named parameters.

omni_playerC

Player actions: teleport, move, look, lookAt, give, say, attack, use, hotbar, drop, sneak, sprint, jump. Per-action fields vary; see the per-action constraints in the tool description.

Inputs: action (required), x, y, z, +11 more.

omni_recipe_validateA

Static linter for a recipe JSON object. Checks the keys the engine actually reads (ModernRecipeRuntime.java:565-1078), the 3x3 pattern limit, the 1.8 vanilla registry for any non-tag ingredient, and the 1.20 result form compatibility.

Inputs: recipe (required), name.

omni_shape_buildingB

One-call house: foundation (optional), floor, walls, roof, centered door, window band. This is the answer to 'build me a house' / 'build me a tower' / 'build me a small castle'.

Inputs: depth (required), width (required), height (required), origin (required), +20 more.

omni_shape_cylinderC

Vertical cylinder. Solid by default. Each y-layer is one or two fill_area runs.

Inputs: center (required), height (required), radius (required), meta, +2 more.

omni_shape_gable_roofB

Gable roof over a footprint. Ridge runs along the given axis. Each course is two fills (both slopes).

Inputs: to (required), from (required), meta, block, +2 more.

omni_shape_hip_roofC

Hip roof over a footprint. Inset by 1 each layer until point. Empty middle kept solid (1 fill at apex).

Inputs: to (required), from (required), meta, block, +1 more.

omni_shape_hollow_boxB

Hollow cuboid shell as up to 6 fill_area ops. Floor/ceiling/walls can be toggled independently.

Inputs: to (required), from (required), meta, block, +3 more.

omni_shape_lineB

3D Bresenham line of place_block ops. Use for paths, fences, wires.

Inputs: to (required), from (required), meta, block.

omni_shape_pyramidB

Stepped square pyramid, 1 fill per layer. hollow=true skips the interior of each layer.

Inputs: center (required), baseRadius (required), meta, block, +1 more.

omni_shape_solid_boxB

Solid cuboid (1 fill_area op). Coordinates are inclusive.

Inputs: to (required), from (required), meta, block.

omni_shape_sphereB

Sphere (or ellipsoid via 3-element radius). Rasterized as x-runs per (y,z) row.

Inputs: center (required), radius (required), meta, block, +1 more.

omni_stateA

Snapshot of the running game: status state, screen, mode, player position/health/food, world, devpatch status. Call this BEFORE every interaction to confirm the world you think is loaded is actually loaded.

Inputs: no named parameters.

omni_world_createB

Create a new world. Optional mods stages them before launch. Template ids: void_single (1x1 indestructible bedrock), void_platform_7x7 (7x7 breakable grass), flat, default.

Inputs: name (required), mods, seed, cheats, +2 more.

omni_world_enterA

Load a saved world by folder name (call omni_worlds first to get the name).

Inputs: name (required).

omni_world_quitA

Back to the main menu. Required before omni_world_create if a world is already running.

Inputs: no named parameters.

omni_world_raycastB

Raycast from origin (or player eye) along direction. Returns block hit, face, distance, plus entity hit if any.

Inputs: origin, distance, direction, includeEntities.

omni_worldsA

List saved worlds. Returns the full list only when no world is running; otherwise reports the running world with a note.

Inputs: no named parameters.

omni_world_scanB

Block and entity snapshot inside a 64^3 box. Hard cap 262,144 blocks. Results truncated to maxBlocks/maxEntities.

Inputs: to (required), from (required), maxBlocks, blockFilter, +3 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.2/5.0

    Scored across 51 tools

    Disambiguation4/5

    Most tools have distinct resource/action pairs, and the high-level prefixes (world_, shape_, map_, mod_, batch_) make the primary purpose clear. However, a few clusters blur together—omni_errors/omni_logs/omni_notifications all address log inspection, and omni_map_guide/omni_map_onboard/omni_map_docs/omni_knowledge are all informational—so an agent may occasionally pick the wrong one. Detailed descriptions save it from a lower score.

    Naming Consistency4/5

    The set consistently uses the omni_ prefix and lowercase snake_case, and the omni_<domain>_<detail> pattern is recognizable across world_, map_, mapdev_, shape_, mod_, and batch_ tools. The pattern is not strictly verb_noun: one-word tools like omni_state, omni_inventory, and omni_chat mix with action-style names like omni_world_create and omni_recipe_validate. This is a minor inconsistency but not chaotic.

    Tool Count1/5

    51 tools is far beyond the 25+ threshold and lands in the 50+ extreme-mismatch range for this rubric. The server covers a broad Minecraft bridge surface, but the sheer number of entries will overwhelm tool-selection and make coherent orchestration harder. Clusters like logs/notifications/errors and the many map_* document tools could be consolidated.

    Completeness4/5

    The tool surface covers the full map-development workflow: world lifecycle, batch validate/apply, shape generation, dev/play mode, mod staging, recipe linting, verification reports, and the map agent context pack. Minor gaps exist—no world delete, no mod list/remove, and no explicit undo/rollback for batches—but those are workaroundable via omni_command or the existing ledger/report machinery. Overall, the core domain is well covered.

    Maintenance

    ActivitySlowing
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      F
      maintenance
      Enables AI-driven interactions with Minecraft through a WebSocket-based server and MCP protocol, allowing external MCP clients and in-game chat to trigger AI tools.
      4
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Provides external AI agents with reliable, observable tools to inspect, navigate, and act in a Minecraft world, enabling autonomous playthroughs such as beating the game.
      18
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables agents to control and observe a Minecraft client through tools for state, entities, commands, chat, recording, waiting, and event replay, bridging any agent runtime to the game.
      MIT