another-houdini-mcp: Instructions file for Codex

AGENTS.md

What This MCP Does

This project provides a Houdini MCP server and plugin that let an agent inspect, build, edit, and validate Houdini scenes and HDAs programmatically.

You can use it to:

• Explore scene structure and node connections
• Create/delete/connect nodes
• Read/set node parameters
• Execute HScript or Python (HOM) when needed
• Inspect local Houdini documentation in help/
• Build validation workflows around geometry and node behavior

Tools Available

Scene and Node Exploration

• get_folder_info
• get_node_info
• get_node_presentation
• get_node_connections
• get_scene_info
• get_sticky_notes

Node Graph Editing

• create_node
• delete_node
• connect_nodes
• remove_connection
• set_parameter

Example and HDA Loading

• install_hda_file
• list_example_nodes
• instantiate_example_asset
• load_example
• instantiate_hda

Node and Parameter Discovery

• list_node_categories
• list_node_types
• get_parameter_info
• get_node_parameters
• get_parameter_overrides
• get_node_documentation

Python/HScript Execution

• execute_python
• execute_hscript

HOM (Python API) Documentation

• list_python_commands
• search_python_documentation
• get_python_documentation

Local Documentation File Tools (help/)

• search_documentation_files
• read_documentation_file

Documentation Expectations

• Before implementing or changing behavior, look up relevant docs.
• Use documentation tools first, then fallback to ad-hoc scripts.
• The project documentation corpus is in help/.

Tutorials Expectations

• Review memory/index.md first to refresh workflow memory.
• Use relevant files in memory/ before inventing a new approach.
• When working in a specific context (for example SOP), refresh context memory first by reading memory/<context>_context.md (for example memory/sop_context.md).
• When working on a specific node, refresh node memory first by reading memory/nodes/<context>/<node>.md if it exists.
• Treat memory/<context>_context.md and memory/nodes/<context>/<node>.md as extended documentation, not activity logs.
• memory/<context>_context.md should contain generalized context-level rules, patterns, heuristics, pitfalls, and cross-node workflows.
• memory/nodes/<context>/<node>.md should contain reusable node-level guidance: intent, setup contracts, parameter interactions, debug tactics, and production usage notes.
• Node-note format should follow memory/node_study_template.md and mirror the richer reference style of memory/nodes/sop/heightfield_erode-2.0.md (Intent, Core Behavior, Key Parameters, Typical Workflow, Production Usage with measured outcomes, Gotchas, Companion Nodes, Study Validation).
• Do not record raw chronological "what I clicked/did" logs in these files; abstract concrete experiences into general lessons.
• If an experiment reveals a failure mode, record the transferable rule (why it failed, how to detect, how to fix), not scene-specific narration.
• Keep session chronology and restart checkpoints in short_term_memory.md only.
• When reviewing an example scene for a target node, inspect all meaningful companion nodes in that example network (not just the target node) and capture reusable patterns in memory notes.
• If companion-node findings are strong enough, add or update their own memory/nodes/<context>/<node>.md files even if they were not the primary study target.
• Do not defer companion-node note updates: while studying one node, proactively edit other node memory notes in the same session when you discover meaningful behavior, patterns, parameters, or edge cases.
• If you struggle with something and then finally solve it, suggest updating the appropriate tutorial in memory/ with that lesson.