Introduction

DCC-MCP-IPC is a multi-protocol IPC adapter layer that bridges DCC (Digital Content Creation) applications — such as Maya, Houdini, Blender, Unreal Engine, and Unity — with the Model Context Protocol (MCP).

It is built on top of dcc-mcp-core (a Rust/PyO3 backend) and exposes DCC functionality as MCP tools via pluggable transport protocols.

What Problem Does It Solve?

AI assistants (Claude, GPT-4, etc.) can use MCP tools to interact with external software. However, DCC applications each have their own scripting APIs, inter-process communication mechanisms, and plugin models. Without a unified adapter, you'd need a separate MCP integration for each DCC.

DCC-MCP-IPC provides:

1. A single API surface — ActionAdapter, SkillManager, transports — regardless of the underlying DCC or protocol.
2. Zero-code skill registration — write a SKILL.md file instead of a full Python MCP tool implementation.
3. Performance — Rust-native IPC transport for latency-sensitive operations; RPyC for DCC-embedded Python; HTTP/WebSocket for REST-based DCCs.

How It Works

AI Assistant (MCP Client)
        ↓
    MCP Server
        ↓
  ActionAdapter  ←──  SkillManager (SKILL.md auto-discovery)
        ↓
  Transport Layer
    ├── RPyC   → Maya / Houdini / Blender (embedded Python)
    ├── IPC    → Rust-native FramedChannel (low-latency)
    ├── HTTP   → Unreal Engine / Unity (REST API)
    └── WS     → WebSocket-based services

The ActionAdapter wraps the Rust ActionRegistry and ActionDispatcher from dcc-mcp-core. Actions are registered with a name, handler function, description, category, and tags. When an MCP tool call arrives, the adapter dispatches it with JSON-serialised parameters.

The SkillManager makes this even simpler: drop a SKILL.md file with frontmatter metadata into a directory, and the SkillScanner automatically registers it as an action. The SkillWatcher watches for file changes and hot-reloads skills without restarting the DCC.

Key Concepts

Concept	Description
Action	A named callable registered with the ActionAdapter. Has description, category, and tags.
Skill	A zero-code action defined by a SKILL.md file. Executed via run.py in the same directory.
Transport	The protocol used to communicate between the MCP server and the DCC.
Adapter	A DCCAdapter wraps a BaseDCCClient to provide a DCC-specific Python API.
Service	A DCCRPyCService or ApplicationRPyCService runs inside the DCC and handles RPC calls.
Discovery	Mechanisms for MCP clients to find running DCC servers (ZeroConf or file-based).

Version 2.0.0 Changes

v2.0.0 is the current unreleased version with significant breaking changes:

• dcc-mcp-core ≥ 0.12.0 required — fully rewritten from Python/Pydantic to Rust/PyO3.
• ActionResultModel.model_dump() → ActionResultModel.to_dict()
• ActionRegistry + ActionDispatcher replace old actions.base / actions.manager modules.
• New Rust-native IPC transport (IpcClientTransport, IpcServerTransport).
• New Skills system (SkillManager, SkillScanner, SkillWatcher).
• Package renamed from dcc-mcp-rpyc → dcc-mcp-ipc.

See the full CHANGELOG for details.

Next Steps

• Installation — Install the package and its dependencies.
• Quick Start — Server + client in 5 minutes.
• Architecture — Deep dive into the component design.