Loading ToolBench
MCP server for Minecraft engineering workspace management, redstone automation, circuit testing, and world interaction via the MC Architect mod
9 tools with domain-specific Minecraft engineering focus. All tools have descriptions (10-200 chars, mostly adequate) and input schemas with enums and typed parameters. However, critical gaps: (1) No output schemas documented anywhere, LLMs cannot predict return structure or chain tools; (2) Parameter descriptions are sparse or missing for many fields (e.g., mc_inspect 'operations' array has no description of structure); (3) Tool names use prefixes (mc_*) rather than action verbs, 'mc_keyboard' doesn't convey 'control' or 'operate'; (4) No error handling guidance in descriptions; (5) No idempotency or retry semantics documented. Schemas are present but incomplete, many optional parameters lack descriptions. This is a specialized domain tool, but definition quality lags production baselines.
Plan incremental deployment against the last completed revision; deploy stages; poll, pause/resume or roll back. Plan/deploy run in background: poll status. Mod advances stages without bridge. Retry deploy after interrupted upload; resume a started assembly. Guards detect external edits.
Begin a reversible spectator camera, move, capture an image artifact, or restore. Game screenshots are evidence separate from the offline geometric preview.
Create an immutable revision by replacing/adding a named module, removing it, importing revised source, or restoring a revision. expectedRevision prevents stale edits. No world writes. Module removal plans air in formerly owned cells; deployment rollback restores before-images.
Read bounded terrain/states/collision/lighting, validate walkable access or compare a small reference. Scans require loaded chunks; deployment loads work chunks automatically.
Optional terminal for an existing physical computer; it does not build or emulate one. Check status.available first: unbound, unloaded or missing hardware returns a reason without live readings, and open returns opened:false. With a valid world-bound profile, enter LDI/ADD/OUT/IN/HLT programs through physical buttons, run, reset, stop or step. Values are 0..255. Check status until operations finish.
No output schemas documented. LLMs cannot predict return structure, field names, or types. Breaks tool chaining and forces agents to guess response format.
Tool names use domain prefix (mc_*) instead of action verbs. 'mc_keyboard' does not convey 'operate', 'control', or 'execute'. LLMs cannot infer intent from name alone.
Many optional parameters lack descriptions. E.g., mc_inspect 'operations' array (maxItems:256) has no description of structure or valid operation types. mc_keyboard 'program' (maxLength:512) has minimal context.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2026-07-28+ | v2 |
Configure opt-in redstone-only acceleration in the current dimension, inspect measured throughput, or disable and return pending ticks to vanilla. World and entity time stay normal. Finish tests and pause construction before changing mode. Parallel wire networks may differ on update-order-sensitive circuits; verify the actual machine.
Run real-game port-based redstone tests: drive existing levers or click vanilla buttons at game ticks, sample signals 0..15, check bit/bus assertions. Tests are deterministic and repeatable; use for validation and regression. Compile a suite from workspace source and a test specification.
Create a persistent engineering workspace from local source JSON; inspect modules/ports/history; read a module; generate an interactive geometry preview. Architecture, machines and redstone use the same model.
Inspect game/world identity, create a flat creative world or open/save-close a local world. Get context before binding a workspace.
No error handling guidance in tool descriptions. No mention of retryability, failure modes, or recovery steps. Agents cannot determine if a failure is transient or fatal.
No idempotency or side-effect semantics documented. Tools like mc_edit, mc_build, mc_test modify state but descriptions do not clarify if they are safe to retry or have irreversible consequences.