dcc-mcp-zbrush

Agent workflow

AI agents should use the shared gateway through dcc-mcp-cli; IDE users may continue to use the MCP endpoint. Prefer typed skills and tools over raw scripts.

Install or update the CLI

dcc-mcp-cli is the preferred control path for every shell-capable agent. If it is missing, ask the user before installing the latest 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"

Keep an official build current through the release manifest:

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

update apply downloads and stages the latest CLI for the next launch. It does not update a running dcc-mcp-server; update that server in its own environment.

dcc-mcp-cli dcc-types
dcc-mcp-cli list
dcc-mcp-cli search --query "<task>" --dcc-type zbrush
dcc-mcp-cli describe <tool-slug>
dcc-mcp-cli call <tool-slug> --json '{"key":"value"}'

dcc-types reports release-catalog support; list reports live sessions. If a tool belongs to an inactive progressive skill, call dcc-mcp-cli load-skill <skill-name> --dcc-type zbrush before retrying. For post-task improvement, attach a stable session id with --meta-json, query dcc-mcp-cli stats --range 24h --session-id <task-id>, then pass the bounded evidence to the review_skill_improvement prompt from dcc-mcp-skills-creator.

ZBrush adapter for the DCC Model Context Protocol ecosystem.

Requires ZBrush 2026.1+ with the official Python SDK (CPython 3.11 embedded in ZBrush).

Showcase

Real ZBrush 2026 evidence from a typed, quiet high-poly workflow: 2,499,970 points / 5,000,000 faces, inspected and exported as the active Dragon subtool. Fantasy Dragon model by Artec 3D, used under CC BY 4.0; the source model is not included in this repository. See the full workflow and copyable prompt.

Quick install

1. Install a fixed wheel version

python -m pip install "dcc-mcp-zbrush==<version>"

The wheel includes the lifecycle CLI and the sidecar bridge payload.

2. Plan and apply the plugin install

Set ZBRUSH_USER_ASSETS_DIR, then run the standard lifecycle. The installer requires an explicit ZBrush path so it can prove the 2026.1+ host floor.

dcc-mcp-zbrush install --version <version> --dcc-path "<ZBrush path>" --python python --dry-run --json
dcc-mcp-zbrush install --version <version> --dcc-path "<ZBrush path>" --python python --yes --json

The fixed release payload is SHA-256 verified before caching. Sidecar mode installs a dedicated module and appends a bounded managed block to the shared Python/init.py; it never overwrites that file. Receipt-backed uninstall and rollback restore the previous state. Applying an install or upgrade stages a candidate and returns exit 50; the candidate is committed only after the exact ZBrush process, start identity, endpoint, version, and loaded module origins pass verify.

See the canonical lifecycle SOP (raw URL).

3. Restart ZBrush

Launch or restart ZBrush, then start the external MCP sidecar:

dcc-mcp-zbrush --mode sidecar --socket-port 9876

The plugin registers a top-level DCC MCP palette through the official ZBrush Python SDK with these actions:

• Copy Instance ID
• Server Info
• About DCC MCP

Instance identity belongs to the MCP server. In embedded mode the first two actions read the public DccServerBase runtime context. The standalone sidecar plugin does not guess identity from environment variables or registry files; it directs you to dcc-mcp-cli list when that external context is not available in the ZBrush process.

4. Verify host readiness

Verify the dynamically allocated instance URL:

dcc-mcp-zbrush verify --version <version> --dcc-path "<ZBrush path>" --python python --json
dcc-mcp-cli list

5. Configure your AI client

Add the MCP server to your AI client config (Cursor, Claude Desktop, etc.):

{
  "mcpServers": {
    "zbrush": {
      "url": "http://127.0.0.1:9765/mcp"
    }
  }
}

How it works

ZBrush does not ship a built-in HTTP REST server. The pre-alpha scaffold that assumed Preferences > Network > Enable HTTP Server was incorrect.

The supported integration paths are:

Rust is not loaded inside ZBrush. The dcc-mcp-core wheel (PyO3) runs in the external sidecar process; importing its extension module into the ZBrush 2026 embedded VM is not a supported runtime path. The ZBrush-facing bridge is Python only and executes requests serially on the host main thread while pumping UI updates between requests. Only one SDK request is admitted at a time; additional requests fail with a retryable busy response, and ping remains available without touching the SDK. A long native operation can still make Windows report ZBrush as not responding because the Maxon SDK call itself is synchronous. If a bridge timeout says the request is still running, do not retry the mutation; poll bridge health until busy becomes false. File import, export, and baking suppress scripted action feedback without wrapping the native operation in zbc.freeze(), so ZBrush can still present progress or a required native dialog.

GoZ C++ SDK is for mesh exchange between DCC apps, not general MCP automation — we do not build the primary adapter on GoZ.

Recommended sidecar mode:

AI Agent → Gateway :9765 → OS-assigned MCP instance → ZBrushMcpServer
         → TCP :9876 → mcp_socket_bridge.py (inside ZBrush) → zbrush.commands

Features (v0.2.0)

• DccServerBase adapter with progressive skill loading
• Bundled skills: zbrush-scripting, zbrush-scene, zbrush-subtool, zbrush-brush, zbrush-viewport, zbrush-interchange, zbrush-import-to-scene
• In-process executor for ZBrush's embedded Python VM
• Optional socket bridge plugin for sidecar deployments
• Top-level DCC MCP palette registered through zbrush.commands
• Gateway election compatible with dcc-mcp-core

Requirements

• ZBrush 2026.1+
• Python 3.10+ on the sidecar host (ZBrush itself ships 3.11)
• dcc-mcp-core >= 0.20.14

Environment variables

Bundled skills

Path concepts

• PYTHONPATH — where Python looks for packages (pip install handles this)
• ZBRUSH_USER_ASSETS_DIR / ZBRUSH_PLUGIN_PATH — plugin scan roots used by ZBrush 2026.1+

pip install dcc-mcp-zbrush puts the Python package on PYTHONPATH. The lifecycle CLI installs host files transactionally; do not copy a bridge over the shared Python/init.py manually.

Skill authoring

Skills lazy-import zbrush.commands and run on the main thread (affinity: main).

from dcc_mcp_core.skill import skill_entry
from dcc_mcp_zbrush.api import import_zbc, with_zbrush, zb_success


@skill_entry
@with_zbrush
def my_tool(**kwargs) -> dict:
    zbc = import_zbc()
    count = zbc.get_subtool_count()
    return zb_success(f"{count} subtool(s)", count=count)

Sidecar mode

After completing the receipt-driven install and restarting ZBrush, run the MCP server outside ZBrush:

The sidecar can also start first: it retains the bridge endpoint and retries the connection on the first tool call after ZBrush becomes available.

dcc-mcp-zbrush --mode sidecar --socket-port 9876

Development

See docs/development.md for source-based setup, testing, and contribution workflow.

References

• ZBrush Python SDK 2026.1
• ZBrush Python environment
• GoZ SDK (mesh exchange only)
• Community reference: newsbubbles/zbrush-mcp (socket bridge pattern)

License

MIT