SideFX Houdini adapter for the DCC Model Context Protocol (MCP) ecosystem — exposes Houdini tools via MCP
Server exposes 6 tools with mixed quality. Project tools (save/load/resume/status) have clear names and basic descriptions but lack detailed parameter documentation and output schemas. CHOP and KineFX tools have better-structured inputs with type info and defaults, but descriptions are terse (apply_filter: 56 chars, apply_mocap: 56 chars, below the 194-char baseline). No tool documents its return schema. Parameter descriptions exist but are minimal; many lack format/constraint guidance. Error handling and recovery paths are not visible in the provided code. Overall: functional but below production baseline.
Apply a CHOP filter (lag, spring, noise, smooth) to a channel
Apply motion capture data to a KineFX rig skeleton
read an existing project.json back
return the rehydration payload an agent needs to restore scene, assets, active skills, tool groups and checkpoint IDs
persist current Houdini project state to .dcc-mcp/project.json
pure-read snapshot of the current state
Output schemas not documented. No tool declares what fields it returns, forcing LLMs to guess downstream field names and types. This breaks tool chaining and wastes tokens on clarification.
Project tool descriptions are under 50 characters and lack context. 'persist current Houdini project state to .dcc-mcp/project.json' is implementation detail, not user intent. Should explain WHEN to call, WHAT it enables, and any prerequisites.
Parameter descriptions lack format/constraint guidance. 'path to project directory' does not specify: absolute vs relative, required file structure, error behavior if missing. LLMs cannot validate inputs without explicit constraints.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
apply_filter and apply_mocap descriptions are terse (56 chars each, vs 194-char baseline). 'Apply a CHOP filter' does not explain when to use it, what filters are available, or what the output represents. Descriptions should be 50-200 chars with clear intent.
No error handling guidance visible. Tools do not document what happens on failure (e.g., invalid filter_type, missing mocap file, unsaved scene). LLMs need recovery paths: 'If mocap file not found, try list_available_mocap_files()' or similar.