maya-mcp-server
MCP server for interacting with Autodesk Maya sessions.
Features
- Multi-session support: Manage multiple Maya sessions from a single MCP server. The server scans for new Maya sessions that have been started and shutdown by the user.
- Full Python expressiveness: Execute arbitrary Python code. Agents can create virtual python modules to expose functions for execution, including by the user.
- Streaming output: Capture stdout/stderr from Maya sessions via MCP resources. Agents can monitor output from their code or user activity.
- Simple Maya-side setup: No modules to install in Maya: leverages Maya's command port to bootstrap itself.
- Easy installation: Install and run via
uvx maya-mcp-server
Installation
# Using uvx (recommended)
uvx maya-mcp-server
# Or install with pip
pip install maya-mcp-server
Usage
Claude Code Configuration
Add to your Claude Code MCP configuration, run:
claude mcp add --transport stdio maya -- uvx maya-mcp-server
The default scope is "local", which adds it to your ~/.claude.json keyed to a particular project directory. Setting --scope=user adds to ~/.claude.json across all projects, and --scope=project to add the configuration into a .mcp.json in the current project directory, so that it can be commited to your repo.
For local development use:
claude mcp add --transport stdio maya -- uv run --directory /path/to/maya-mcp-server/ maya-mcp-server
Maya Setup
The server automatically discovers Maya sessions via command ports. To enable a Python command port in Maya:
import maya.cmds as cmds
cmds.commandPort(name=":7002", sourceType="python")
Or add to your userSetup.py for automatic startup.
Tools
Tools accept an optional session_key parameter for targeting specific sessions.
If only one Maya session exists, it will be auto-selected.
Session keys are returned by list_sessions and add_session.
| Tool | Description |
|------|-------------|
| list_sessions | List all active Maya sessions. Returns session info including session_key for use with other tools/resources. |
| add_session | Manually add a Maya session at a specific host:port. Use when auto-discovery doesn't find your session. |
| write_module | Create a virtual Python module in Maya. Useful for defining reusable functions. |
| execute_code | Execute Python code in a session. Supports result capture modes: NONE, JSON, RAW. |
Resources
| Resource | Description |
|----------|-------------|
| maya://sessions/{session_key}/info | Session information (pid, user, maya_version, scene_name, scene_path) |
| maya://sessions/{session_key}/output | Captured stdout/stderr output from the session |
Similar tools
This looks to be the first publicly available MCP server for Maya and I was inspired by a few aspects of this tool, especially the goal of zero Maya-side setup.
Disadvantages:
- The MCP server is bound to a single Maya session running on the default port.
- It is limited to a bespoke set of tools. This could be seen as a security advantage, but it cripples the ability of an agent to do just about anything.
- No support for reading stdout or stderr, so the agent is blind to what's happening in the Maya session.
- Less robust approach to capturing command output (e.g. does not check if code is indented within a
for loop or function)
- Can't run via
uvx, or pip install from pypi.
This is the original LLM integration for Maya, which embeds ChatGPT directly in a PySide window and enables the LLM to respond to user commands and queries by executing code in the session.
Disadvantages:
- Not an MCP server, so it cannot take full advantage of agentic workflows.
- Only works with ChatGPT.
... View full README on GitHub
