DCC-MCP-CoreMCP + Skills for DCC AI

Production-grade foundation combining Model Context Protocol (MCP) and a zero-code Skills system. Connect AI to Maya, Blender, Houdini, Photoshop — without context explosion.

Get Started

Why MCP + Skills?

GitHub

🎯

Solves MCP Context Explosion

Session isolation pins each AI session to one DCC instance. tools/list returns 150 tools instead of 750. Progressive discovery by DCC type, scope, and product.

🔌

Zero-Code Skill Registration

Write SKILL.md + scripts/ → instant MCP tools. No Python glue code needed. Supports Python, MEL, MaxScript, Bash, PowerShell, and more.

🏆

Version-Aware Gateway Election

Multiple DCC instances compete for gateway role. Newest version automatically takes over. No manual failover — just semantic version comparison.

🦀

Rust-Powered Core

Zero runtime Python dependencies. Zero-copy IPC via Named Pipes and Unix Sockets. rmp-serde serialization. LZ4 shared memory. Sub-millisecond tool calls.

🔒

SkillPolicy & Scope

allow_implicit_invocation controls whether AI can call a skill without explicit load_skill. products filter visibility by DCC. Trust levels (repo < user < system < admin).

📦

Instance Tracking

Every DCC registers pid, display_name, active scene, and open documents. AI agents route to the right instance by document or display name.

🛡️

Sandbox & Audit Log

Policy-based access control with in-memory audit logging. Define what AI can and cannot do per DCC type.

🌐

MCP Streamable HTTP

Built-in server (2025-03-26 spec). Claude Desktop, Cursor, and other MCP clients connect directly over HTTP. Background thread, never blocks DCC.

📊

Structured Results

Every tool returns (success, message, context, next_steps). AI agents reason clearly about outcomes. No fragile text parsing.