dcc-mcp-3dsmax

3ds Max plugin for the DCC Model Context Protocol (MCP) ecosystem.

Agent workflow

Choose the connection path that matches your client:

• Shell or headless agent: use dcc-mcp-cli through the shared gateway.
• IDE MCP client: connect to http://127.0.0.1:9765/mcp.
• 3ds Max setup: follow install.md to install the runtime and startup hook.

Prefer typed skills and tools over raw scripts.

Install and update the CLI

If dcc-mcp-cli is missing, obtain user consent before installing the official release:

# Linux/macOS
curl -fsSL https://raw.githubusercontent.com/dcc-mcp/dcc-mcp-core/main/scripts/install-cli.sh | sh

# Windows PowerShell
powershell -ExecutionPolicy Bypass -c "irm https://raw.githubusercontent.com/dcc-mcp/dcc-mcp-core/main/scripts/install-cli.ps1 | iex"

Check for updates, then stage the latest CLI for the next launch:

dcc-mcp-cli update check
dcc-mcp-cli update apply

update apply stages the CLI only. It does not update a running dcc-mcp-server; update that server in its own environment.

Features

• Sidecar MCP Server: Starts dcc-mcp-server.exe sidecar and keeps 3ds Max scene edits on the main thread
• Progressive Skill Loading: Discover skills without loading them immediately
• Verified Undo/Redo: 3dsmax-undo reverses host operations and confirms each step, and every destructive tool declares whether it is undoable — see docs/UNDO.md
• Shared Gateway: Registers with the stable gateway at http://127.0.0.1:9765/mcp
• Job Persistence: SQLite-based job storage for long-running operations
• Prometheus Metrics: Optional /metrics endpoint for monitoring

Agent install (recommended)

The fastest way to install is to let your AI agent do it. In Cursor, Claude, or any MCP-capable agent host, ask:

Please install and verify dcc-mcp-3dsmax.

First clone or locate the dcc-mcp-3dsmax repository and resolve its absolute
path as REPO_ROOT. Then read these absolute-path files before making changes:

- <REPO_ROOT>\install.md
- <REPO_ROOT>\skills\dcc-mcp-3dsmax-setup\SKILL.md

Follow the Install SOP v1 lifecycle exactly:

1. Locate my 3dsmaxpy.exe. If it cannot be found, ask me for the absolute path.
2. Run `dcc-mcp-3dsmax install --json --dry-run` with explicit `--dcc-path`
   and `--python`, inspect the machine-readable plan, then rerun with `--yes`.
3. Preserve the receipt and follow any restart next step.
4. Run `dcc-mcp-3dsmax verify --json`; only
   `verify.directly_usable=true` proves typed tools are callable.
5. Configure the MCP host for http://127.0.0.1:9765/mcp only after verification.

The agent reads install.md, uses the standard lifecycle verbs, and preserves the structured receipt. The drag-and-drop MZP remains documented as the manual release-artifact path. Repository CI is hermetic and does not claim that a licensed 3ds Max host was opened.

Installation

pip install dcc-mcp-3dsmax

Quickstart (inside 3ds Max MAXScript Listener)

import dcc_mcp_3dsmax

# Start on a random instance port; the public gateway stays fixed.
server = dcc_mcp_3dsmax.start_server()

# Progressive loading — discover skills without loading them immediately
n = server.discover_skills()        # scan paths, register tool metadata
server.load_skill("3dsmax-scene")  # lazy-load a specific skill

dcc_mcp_3dsmax.stop_server()

Runtime Bridge

Start the runtime bootstrap inside 3ds Max:

python.ExecuteFile @"C:\path\to\dcc-mcp-3dsmax\examples\start_sidecar_bridge.py"

This starts the agent-callable embedded MCP runtime, registers bundled 3ds Max skills/tools, and routes main-affinity scene edits through the shared dcc-mcp-core 0.20.22 UI dispatcher, dcc-cua 0.4 UI Control, pump abstractions, gateway guardian, dynamic instance ports, and six-state readiness probe (process, dispatcher, dcc, skill_catalog, host_execution_bridge, main_thread_executor). The legacy random-port /dispatch bridge remains available for local diagnostics. The 3ds Max instance is registered with the stable gateway at http://127.0.0.1:9765/mcp. See docs/SIDECAR.md and the bundled skill index in docs/BUNDLED_SKILLS.md.

The bootstrap also installs a DCC MCP menu with Start Server, Stop Server, Open Gateway Admin, and Print Status commands. 3ds Max shutdown triggers the same stop path via #preSystemShutdown, so the runtime and local bridges are cleaned up when the host exits.

Skill Development

Create a skill with SKILL.md metadata file and Python scripts:

# my_skill/action_create_box.py
from dcc_mcp_3dsmax.api import max_success, with_max

@with_max
def main(width: float = 100.0, height: float = 100.0, depth: float = 100.0) -> dict:
    import pymxs
    rt = pymxs.runtime

    box_obj = rt.Box(width=width, height=height, depth=depth)
    return max_success("Created box", object_name=str(box_obj))

Environment Variables

Normal MZP installs do not need DCC_MCP_3DSMAX_PORT or DCC_MCP_GATEWAY_PORT: the adapter uses an internal ephemeral MCP port and the shared gateway default.

For Rez-style deployment, launch 3ds Max with package roots in the environment instead of copying packages into the user scripts folder. A pipeline package cache can expose paths such as <package-cache>/dcc_mcp_core, <package-cache>/dcc_mcp_3dsmax, and <package-cache>/dcc_mcp_server through the root variables above. The MZP installer uses isolated version directories under <user scripts>/dcc_mcp_3dsmax/versions/, so installing a new payload does not delete the version that may already be loaded by the running 3ds Max process. After install, the active payload is added to sys.path, obsolete payload directories are cleaned up where possible, and the runtime is started. The generated startup script repeats that activation on future 3ds Max launches, so the bridge comes back automatically after restart. Uninstall cleanup uses the import-light lifecycle helpers from dcc-mcp-core when available, so locked native files are reported as restart-required instead of disappearing into a generic PermissionError.

Requirements

• 3ds Max 2017 or later (Python 3.x with pymxs support)
• Python >= 3.7
• dcc-mcp-core >= 0.20.24
• dcc-mcp-server >= 0.20.22

License

MIT