Loading ToolBench
MCP server for simulating interactive world steps with neural rendering, voxel physics, and dynamic lighting
Critical failures across all dimensions. The single tool 'execute_skill_action' has a minimal description (67 chars, below the 194-char baseline), lacks output schema documentation entirely, and parameters lack proper type constraints. The mcp_server.py implementation is a stub that prints JSON and exits, it does not actually implement the MCP protocol. Tool definitions are inferred from client.py, not explicitly registered via MCP mechanisms. No error handling, no parameter validation, no security considerations.
Execute a skill action in the realtime interactive world model neural simulator
No output schema documented. The tool returns a complex object with 7 fields (world_frame_step_id, action_applied, rendering_framerate_fps, latent_diffusion_step_latency_ms, voxel_physics_consistency_score_pct, dynamic_lighting_propagated, h265_video_frame_stream_url) but LLMs have no specification of these fields, types, or when they are present.
Tool description is only 67 characters and lacks actionable context. It does not explain WHEN to use this tool, what the action_vector format actually accepts, what the context_frame_buffer_size controls, or what the returned fields mean. Baseline is 194 chars.
Parameter 'user_action_vector' is a free-form string with no enum constraints or format specification. The description mentions examples like 'MOVE_FORWARD_JUMP_MINE_BLOCK' but does not define the grammar, valid tokens, or how to compose them. LLMs will hallucinate invalid action vectors.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 25 | 2026-07-28+ | v2 |
No error handling or recovery guidance. If the simulator fails, times out, or rejects an action, the tool provides no error classification, actionable message, or retry strategy. LLMs cannot self-correct.
Tool definitions are inferred from client.py, not explicitly registered via MCP protocol. The mcp_server.py is a non-functional stub that only prints a JSON status message and exits.