DCC-MCP-IPC
Multi-protocol IPC adapter layer for integrating DCC software (Maya, Houdini, Blender, Unreal, Unity) with MCP, enabling tool registration and remote control across multiple transport protocols.
README
DCC-MCP-IPC
<div align="center"> <img src="https://raw.githubusercontent.com/loonghao/dcc-mcp-ipc/main/logo.svg" alt="DCC-MCP-IPC Logo" width="200"/>
Multi-protocol IPC adapter layer for DCC software integration with Model Context Protocol (MCP).
Built on top of dcc-mcp-core (Rust/PyO3 backend), it provides a high-performance, type-safe framework for exposing DCC functionality as MCP tools across multiple transport protocols.
Documentation: docs site | v2.0.0 (Unreleased) — Breaking changes ahead, see CHANGELOG.md
Why DCC-MCP-IPC?
| Feature | Description |
|---|---|
| Protocol-agnostic | RPyC for embedded-Python DCCs (Maya/Houdini/Blender), HTTP for Unreal/Unity, WebSocket, and Rust-native IPC for maximum throughput. |
| Zero-code Skills | Drop a SKILL.md file into a directory — SkillManager auto-registers it as an MCP tool. No Python boilerplate needed. |
| Rust-powered core | Action dispatch, validation, and telemetry handled by dcc-mcp-core via PyO3; Python layer focuses on DCC-specific glue code. |
| Hot-reload skills | SkillWatcher monitors skill directories and re-registers tools on file changes without restarting the DCC. |
| Service discovery | ZeroConf (mDNS) + file-based fallback for automatic DCC server detection. |
| Connection pooling | ConnectionPool with auto-discovery for efficient client-side connection reuse. |
Features
- Thread-safe RPyC / HTTP / WebSocket / Rust-native IPC server implementations
- Skills system — zero-code MCP tool registration from
SKILL.mdfrontmatter with hot-reload - Action system backed by
ActionRegistry+ActionDispatcher(Rust) — JSON-serialised parameter dispatch - Transport factory — pluggable transport layer (
rpyc,http,websocket,ipc) - Service discovery: ZeroConf (mDNS) + file-based fallback via
ServiceDiscoveryFactory - Async client (
asyncio) for non-blocking operations - Abstract base classes for creating DCC-specific adapters (
DCCAdapter) and services - Application adapter pattern for generic app integration
- Scene & snapshot interfaces over RPyC and HTTP transports
- Mock DCC services for testing without actual DCC applications
- Comprehensive error handling with custom exception hierarchy
Architecture
graph TD
A[AI Assistant / MCP Client] --> B[MCP Server Layer]
B --> C[ActionAdapter\nActionRegistry + ActionDispatcher]
C --> D[RPyC Transport\nDCC Python env]
C --> E[IPC Transport\nRust FramedChannel]
C --> F[HTTP Transport\nUnreal / Unity REST]
C --> G[WebSocket Transport]
G[SkillManager\nSKILL.md auto-discovery] --> C
H[dcc-mcp-core\nRust/PyO3 backend] --> C
D --> I[DCC App\nMaya / Houdini / Blender]
E --> I
F --> J[DCC App\nUnreal / Unity]
style C fill:#e1f5fe
style H fill:#fff3e0
style G fill:#e8f5e9
Key Components
| Component | Module | Description |
|---|---|---|
ActionAdapter |
action_adapter.py |
Wraps Rust ActionRegistry + ActionDispatcher; registers handlers and dispatches JSON-parameterised calls |
SkillManager |
skills/scanner.py |
Scans directories for SKILL.md skills, registers them as action handlers, supports hot-reload |
DCCServer |
server/dcc.py |
Manages the RPyC/IPC server lifecycle inside a DCC process |
BaseDCCClient |
client/base.py |
Core client connection/call logic with auto-discovery |
ConnectionPool |
client/pool.py |
Connection pooling for efficient resource management |
IpcClientTransport / IpcServerTransport |
transport/ipc_transport.py |
Rust-native framed-channel IPC, registered as "ipc" protocol |
ServiceDiscoveryFactory |
discovery/factory.py |
Strategy-pattern selector for ZeroConf or file-based discovery |
MockDCCService |
testing/mock_services.py |
Simulates DCC applications for testing |
Installation
pip install dcc-mcp-ipc
With optional ZeroConf support:
pip install "dcc-mcp-ipc[zeroconf]"
Or with Poetry:
poetry add dcc-mcp-ipc
Requirements
- Python >= 3.8 (< 4.0)
- dcc-mcp-core >= 0.12.0 (< 1.0.0)
- rpyc >= 6.0.0 (< 7.0.0)
- Optional: zeroconf >= 0.38.0 for mDNS discovery
Quick Start
Server-side (within DCC application)
<augment_code_snippet path="README.md" mode="EXCERPT">
from dcc_mcp_ipc.server import create_dcc_server, DCCRPyCService
class MayaService(DCCRPyCService):
def get_scene_info(self):
return {"scene": "Maya scene info"}
def exposed_execute_cmd(self, cmd_name, *args, **kwargs):
pass
server = create_dcc_server(
dcc_name="maya",
service_class=MayaService,
port=18812,
)
server.start(threaded=True)
</augment_code_snippet>
Client-side
<augment_code_snippet path="README.md" mode="EXCERPT">
from dcc_mcp_ipc.client import BaseDCCClient
client = BaseDCCClient("maya", host="localhost", port=18812)
client.connect()
result = client.call("get_scene_info")
client.disconnect()
</augment_code_snippet>
Usage Guide
Action System (v2.0.0+)
The Action system is built on the Rust-backed ActionRegistry + ActionDispatcher. All parameters are JSON-serialised:
<augment_code_snippet path="README.md" mode="EXCERPT">
from dcc_mcp_ipc.action_adapter import ActionAdapter, get_action_adapter
adapter = get_action_adapter("maya")
def create_sphere(radius: float = 1.0, name: str = "sphere1") -> dict:
return {"success": True, "message": f"Created {name}", "context": {"name": name}}
adapter.register_action(
"create_sphere",
create_sphere,
description="Create a sphere primitive",
category="modeling",
tags=["primitive", "mesh"],
)
result = adapter.call_action("create_sphere", radius=2.0, name="mySphere")
print(result.success) # True
print(result.to_dict()) # {"success": True, ...}
</augment_code_snippet>
Zero-code Skills via SkillManager
Drop a SKILL.md file into a directory structure:
my_skills/
create_light/
SKILL.md # frontmatter: name, description, tools, scripts
run.py # executed when the tool is called
<augment_code_snippet path="README.md" mode="EXCERPT">
from dcc_mcp_ipc.skills import SkillManager
from dcc_mcp_ipc.action_adapter import get_action_adapter
adapter = get_action_adapter("maya")
mgr = SkillManager(adapter=adapter, dcc_name="maya")
mgr.load_paths(["/pipeline/skills"])
mgr.start_watching()
# Now "create_light" is callable as an MCP tool
result = adapter.call_action("create_light", intensity=100.0)
</augment_code_snippet>
Connection Pool
<augment_code_snippet path="README.md" mode="EXCERPT">
from dcc_mcp_ipc.client import ConnectionPool
pool = ConnectionPool()
with pool.get_client("maya", host="localhost") as client:
result = client.call("execute_cmd", "sphere", radius=5)
print(result)
</augment_code_snippet>
Service Factories
Three factory patterns are available for different lifecycle needs:
<augment_code_snippet path="README.md" mode="EXCERPT">
from dcc_mcp_ipc.server import (
create_service_factory,
create_shared_service_instance,
create_raw_threaded_server,
)
class SceneManager:
def __init__(self):
self.scenes = {}
def add_scene(self, name, data):
self.scenes[name] = data
scene_manager = SceneManager()
# Per-connection instance
service_factory = create_service_factory(MayaService, scene_manager)
# Shared singleton instance
shared_service = create_shared_service_instance(MayaService, scene_manager)
# Raw threaded server
server = create_raw_threaded_server(service_factory, port=18812)
server.start()
</augment_code_snippet>
Rust-native IPC Transport
Zero-copy low-latency messaging via the Rust core:
<augment_code_snippet path="README.md" mode="EXCERPT">
import os
from dcc_mcp_core import TransportAddress
from dcc_mcp_ipc.transport.ipc_transport import (
IpcClientTransport,
IpcServerTransport,
IpcTransportConfig,
)
# Client side
config = IpcTransportConfig(host="localhost", port=19000)
transport = IpcClientTransport(config)
transport.connect()
result = transport.execute("get_scene_info")
transport.disconnect()
# Server side (inside DCC plugin)
def handle_channel(channel):
msg = channel.recv()
# process and respond ...
addr = TransportAddress.default_local("maya", os.getpid())
server = IpcServerTransport(addr, handler=handle_channel)
bound_addr = server.start()
</augment_code_snippet>
Creating a Custom DCC Adapter
<augment_code_snippet path="README.md" mode="EXCERPT">
from dcc_mcp_ipc.adapter import DCCAdapter
from dcc_mcp_ipc.client import BaseDCCClient
class MayaAdapter(DCCAdapter):
def _initialize_client(self) -> None:
self.client = BaseDCCClient(
dcc_name="maya",
host=self.host,
port=self.port,
connection_timeout=self.connection_timeout,
)
def create_sphere(self, radius: float = 1.0):
self.ensure_connected()
assert self.client is not None
return self.client.execute_dcc_command(f"sphere -r {radius};")
</augment_code_snippet>
Async Client
<augment_code_snippet path="README.md" mode="EXCERPT">
import asyncio
from dcc_mcp_ipc.client.async_dcc import AsyncDCCClient
async def main():
client = AsyncDCCClient("maya", host="localhost", port=18812)
await client.connect()
result = await client.call("get_scene_info")
await client.disconnect()
asyncio.run(main())
</augment_code_snippet>
Testing with Mock Services
<augment_code_snippet path="README.md" mode="EXCERPT">
import threading
from dcc_mcp_ipc.testing.mock_services import MockDCCService
from dcc_mcp_ipc.client import BaseDCCClient
server = MockDCCService.start(port=18812)
client = BaseDCCClient("mock_dcc", host="localhost", port=18812)
client.connect()
info = client.get_dcc_info()
print(info) # {"name": "mock_dcc", ...}
client.disconnect()
server.stop()
</augment_code_snippet>
Development
Setup
git clone https://github.com/loonghao/dcc-mcp-ipc.git
cd dcc-mcp-ipc
poetry install
Running Tasks
nox -s pytest # Run tests
nox -s lint # Lint (mypy + ruff + isort)
nox -s lint-fix # Auto-fix lint issues
nox -s build # Build distribution packages
Project Structure
dcc-mcp-ipc/
├── src/dcc_mcp_ipc/
│ ├── __init__.py # Lazy-import public API surface
│ ├── action_adapter.py # Action system (Rust-backed)
│ ├── adapter/ # DCC & application adapters
│ ├── client/ # Synchronous & async clients + pool
│ ├── server/ # RPyC server + factories + lifecycle
│ ├── transport/ # RPyC / HTTP / WS / IPC transports
│ ├── discovery/ # ZeroConf + file-based service discovery
│ ├── skills/ # SkillManager zero-code system
│ ├── scene/ # Scene operations interface
│ ├── snapshot/ # Snapshot interface
│ ├── application/ # Generic application adapter/service/client
│ ├── testing/ # Mock services for testing
│ └── utils/ # Errors, DI, decorators, RPyC helpers
├── tests/ # 68 test files mirroring source layout
├── examples/ # Usage examples
├── docs/ # VitePress documentation site
└── nox_actions/ # Nox task definitions
License
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.