Typed operations exposed to compatible AI clients.
01 / Agentic tool infrastructure / 2026
MATERIAL PILOT
A local-first MCP bridge that lets AI agents inspect, create, validate, and export procedural material graphs.
- Role
- Architecture, implementation, packaging, testing, and documentation
- Duration
- July 2026
- Status
- open-source
Generated procedural-node definitions available to the bridge.
Read-only, assisted, workspace-autonomous, and developer.
The bridge operates without external data transmission.
A / EVIDENCE
REAL OUTPUTS, RELEASES, AND SOURCE.
Product captures link to the live or published source. The architecture view is generated only from documented project layers and is not presented as a product screenshot.
SOURCE ↗EVIDENCE / 01
MaterialPilot's public GitHub repository, including the package structure and documented project scope.
B / PROBLEM
What needed to change
Procedural material workflows are visual, stateful, and difficult for language-model clients to modify safely. Agent actions need a typed interface, explicit permissions, and revision-aware editing.
C / SYSTEM RESPONSE
What I built
MaterialPilot exposes Material Maker through an MCP-compatible TypeScript server and an authenticated loopback GDScript bridge, with transactional graph operations and rollback support.
D / ARCHITECTURE
THE SYSTEM, LAYER BY LAYER.
Each layer below names its responsibility directly, preserving the distinction between repository evidence and explanatory architecture.
AI client
Plans work and invokes typed MCP operations.
TypeScript server
Validates requests, permissions, and expected revisions.
WebSocket JSON-RPC
Carries authenticated loopback messages.
GDScript bridge
Applies atomic graph changes inside Material Maker / Godot.
E / ENGINEERING DECISIONS
Revision-safe mutation
Expected-revision checks and snapshot-first execution protect graph integrity when multiple actions occur.
TRADE-OFFAdditional validation work is exchanged for predictable failure and recovery.
Permission-gated autonomy
Four permission modes let users choose how much authority an agent receives.
TRADE-OFFMore modes increase setup surface but keep sensitive actions explicit.
Generated node contracts
Generated definitions provide a broad typed surface without hand-maintaining every node.
TRADE-OFFGeneration must remain synchronized with supported Material Maker versions.
F / STACK
TypeScriptNode.jsMCPGDScriptGodot 4.7WebSocket JSON-RPC
G / LIMITATIONS
WHERE THE SYSTEM STOPS.
- Requires a local Material Maker / Godot environment; no rendered material exports are committed, so this portfolio shows source evidence and documented architecture rather than inventing product output.
- Compatibility follows documented packages for Windows, Linux, and macOS and still depends on host application behavior.
- Real material output quality remains dependent on the graph instructions provided to the system.
H / WHAT I LEARNED
WHAT THE BUILD CLARIFIED.
- Agent tooling benefits from the same transactional guarantees expected in conventional developer infrastructure.
- Explicit permission modes make autonomy easier to reason about and document.