MCP for Maya — Connect AI Agents to Maya
Maya scenes are graphs, references and namespaces before they are shapes. Agents work against that structure — resolving names, respecting references, and never guessing which node they just rewired.
- Host software
- Maya 2024, 2025 and 2026
- Runtime
- Python 3.10+ (mayapy) and the Maya Python API 2.0
- Transport
- stdio (local) and streamable HTTP (studio)
- Install
- Maya module (mod) + npx dccmcp-maya
- Modes
- Interactive session, mayapy standalone, render-farm workers
- Platforms
- Windows, macOS, Linux
- Last reviewed
- 2026-09-20
What is MCP for Maya?
MCP for Maya is a local MCP server plus a Maya plugin that connects AI agents to a real Maya session. Agents discover 10 typed tools — including scene.inspect, namespace.resolve, node.create and node.connect — and call them to inspect, create, modify and validate work inside the live application.
Unlike generated Maya scripts, every call is validated against your current document, classified as read, write or execute, and recorded in an audit log. Maya stays open, the document stays live, and destructive operations are checkpointed so they can be rolled back.
It works with Claude Code, Codex, Cursor, GitHub Copilot, Claude Desktop, Gemini CLI, Trae, Doubao, Qwen Code, CodeBuddy, Custom MCP clients over stdio or streamable HTTP.
- Host software: Maya 2024, 2025 and 2026
- Runtime: Python 3.10+ (mayapy) and the Maya Python API 2.0
- Transport: stdio (local) and streamable HTTP (studio)
What goes wrong when agents drive Maya without a contract
Namespaces quietly break every hardcoded name
A script that works in the artist's session fails on the shot file because the reference came in as props_v003:pCube1. Agents need to resolve names before they use them, not after a traceback.
One bad node graph edit corrupts the rig
Rewiring a dependency graph connection or overwriting a driven key silently breaks a rig that took a week to build — and the damage only shows up in the next animation pass.
Heavy scenes punish exploratory automation
Agents that open, evaluate and rebuild the whole scene to answer a question are unusable on production shots. Reads have to be scoped and cheap.
How MCP for Maya keeps intent, structure and traceability intact
Namespace-aware resolution
Tools accept short or full names and resolve them against the live namespace tree, reporting the resolved path so the agent knows exactly which node it touched.
Graph edits with preview
Connection and attribute changes are returned as a diff before they are applied, and every write runs inside a checkpointed transaction.
Reference-safe assembly
The agent can inspect references and their versions, swap them deliberately, and is blocked from editing referenced content unless the policy allows it.
mayapy batch parity
The same tool surface runs in mayapy standalone for shot-wide operations, so an agent can apply a fix across a sequence without opening the UI.
MCP for Maya tools — every tool and its risk class
scene.inspectScoped snapshot of the DAG, namespaces, references, layers and render settings.readnamespace.resolveResolve short, ambiguous or wildcard names against the live namespace tree.readnode.createCreate DAG or DG nodes with validated parenting and attribute presets.writenode.connectWire and rewire dependency graph connections with a returned diff.writeattribute.setSet transform or custom attributes, including driven-key keys.writereference.manageList, load, unload and version-swap references without touching referenced edits.writeanimation.keyframeRead and write keyframes and tangents across selected channels.writecurve.editEdit animation curves: scale, offset, retime and bake with explicit ranges.writerender.submitSubmit renders with declared camera, range and renderer settings.executescene.checkpointCreate, list and restore named scene checkpoints.write
Install MCP for Maya in four steps
Pre-release. These commands describe the interface we are shipping for MCP for Maya — the packages are not published yet. Get notified when MCP for Maya launches.
Install the Maya module
Drop the DCCMCP module into your Maya module path. Per-project module directories are supported for studios that keep tooling inside a pipeline repo.
cp -R dccmcp-maya/ /usr/local/maya/modules/
Start the bridge
Run the bridge against the open session, or let it launch mayapy for batch work.
npx dccmcp-maya@latest serve --session active
Register the client
Register the server with Claude Code, Codex, Cursor, OpenClaw or any other MCP client. Studio builds can point at the shared HTTP transport instead.
{
"mcpServers": {
"maya": { "command": "npx", "args": ["dccmcp-maya", "serve"] }
}
}Protect referenced content
Deny writes into referenced nodes by default so an agent cannot silently modify shared assets.
dccmcp policy deny --tool node.create --when referenced=true
MCP for Maya works with the agents you already use
- Claude Code
- Codex
- Cursor
- GitHub Copilot
- Claude Desktop
- Gemini CLI
- Trae
- Doubao
- Qwen Code
- CodeBuddy
- Custom MCP clients
Any other MCP-compatible client works too — one config entry, stdio by default.
Other MCP integrations that pair with Maya
10 integrations available
Compare tool lists, support matrices and policies for every supported application.
Questions teams ask before they install
Maya 2024, 2025 and 2026 are supported on the beta track, using the Python 3 runtime and the Maya Python API 2.0. Python 2 / Maya 2022 and older are available to enterprise customers only.
Graph edits and attribute writes are checkpointed and returned as a diff, referenced content is read-only by default, and a policy can deny specific node types outright — for example, anything under a rig's control group.
Yes. The same tools are exposed in mayapy standalone, which is how most teams apply an approved change across an entire shot list or hand work to a render farm.
References are first-class: the agent can inspect them, check versions and swap them deliberately. USD stages can be read and, with an approving policy, edited through Maya's USD integration.
Render submission is renderer-agnostic — the agent declares the renderer and settings, and the server validates them against what is actually installed in your session. Arnold, V-Ray and RenderMan are all used in our test corpus.
Put an agent inside your real tools — without putting your pipeline at risk
Install an integration, set a policy, and let your agent inspect, build and validate work in Blender, Maya, Houdini, 3ds Max, Rhino, ZBrush, Photoshop, FreeCAD, QGIS or OpenCV. Free to start, local by default.
macOS · Windows · Linux — no cloud dependency required