Common AI + DCC tasks

DCC-MCP is the typed control plane between an agent and creative software. For these tasks, start with the public dcc-mcp Skill; it teaches the agent to discover the right tool instead of inventing scripts or guessing tool names.

How do I control Maya with AI?

Install the dcc-mcp Agent Skill, connect the official Maya adapter, and use dcc-mcp-cli to discover and call typed Maya tools. The agent should inspect the machine first and ask before installing the CLI or adapter.

Set up DCC-MCP so you can control Maya on this machine. Install and follow the public @loonghao/dcc-mcp Skill. Inspect the current CLI and Maya adapter first; ask before installing or changing system state. Open or detect Maya, verify the gateway and live Maya instance, perform one read-only search and describe flow, then report the connected instance, available capability, and next safe Maya action. Do not modify my current scene during setup.

Useful evidence:

dcc-mcp-cli health
dcc-mcp-cli list
dcc-mcp-cli search --query "inspect Maya scene" --dcc-type maya

How do I control Blender with AI?

Use the same Skill and CLI with the official Blender adapter. DCC-MCP keeps the workflow consistent while Blender owns its host-specific installation and typed tools.

Set up DCC-MCP so you can control Blender on this machine. Install and follow the public @loonghao/dcc-mcp Skill. Inspect the current CLI and Blender adapter first; ask before installing or changing system state. Open or detect Blender, verify the gateway and live Blender instance, perform one read-only search and describe flow, then report the connected instance, available capability, and next safe Blender action. Do not modify my current scene during setup.

Useful evidence:

dcc-mcp-cli health
dcc-mcp-cli list
dcc-mcp-cli search --query "inspect Blender scene" --dcc-type blender

How do I create ten random spheres in Maya?

Give the result you want to the Agent Skill. It will search the live Maya tool catalog, follow the returned next_step, call typed tools, and validate the scene. The prompt should define randomness, bounds, naming, and acceptance instead of prescribing an unverified tool slug.

Use the dcc-mcp Skill to connect to my live Maya instance. In a new group named ai_random_spheres, create exactly 10 polygon spheres with deterministic seed 42. Give each sphere a unique name and place it at a random position inside X -10..10, Y 0..10, Z -10..10 while avoiding visible intersections. Do not delete or replace existing scene content. Discover the typed Maya tools first, follow every returned next_step, then validate the final sphere count, names, transforms, and group membership. Show me the evidence and do not save the scene unless I ask.

I want to make a game. Where do I start?

Start with the smallest playable loop, then choose the connected engine: Unreal Engine, Unity, or Godot. Use the Marketplace for optional, license-aware assets and the Showcase for proven prompt patterns.

Help me make a small playable game with DCC-MCP. First ask me for the engine (Unreal, Unity, or Godot), target platform, visual style, and one-sentence playable loop. Reduce the idea to one level and one win or fail condition. Inventory the connected DCCs and engine, discover typed tools, and inspect the official Marketplace for any license-safe assets we need. Ask before installing packages, downloading assets, or changing system state. Build the smallest playable slice, run it, validate controls and the win/fail loop, capture evidence, and report the project and packaged-build paths. Do not claim completion without a runnable build.

How do I edit photos with AI?

Use the official Photoshop adapter for retouching, color correction, masks, layers, and export. The agent should preserve the original, work non-destructively, and ask for the intended look and output format before editing.

Use the dcc-mcp Skill to connect to Photoshop and help me edit this image. First ask what must change, what must remain untouched, the reference look, and the delivery format. Preserve the original document, use new layers, adjustment layers, masks, or Smart Objects where supported, and never flatten or overwrite the source without my approval. Discover the typed Photoshop tools, make a bounded first pass, compare the result with the brief, validate dimensions, color mode, layer structure, and missing assets, then export to a new path only after I approve the preview.

How do I edit or composite a film with AI?

Route the job by intent: Premiere Pro for timeline editing, After Effects for motion graphics and layer-based shot work, or Nuke for node-based compositing. Do not move a workflow to another host merely because its tool catalog is easier to reach.

Use the dcc-mcp Skill to help me edit or composite this film. First ask whether the task is timeline editing in Premiere, motion graphics or layer compositing in After Effects, or node compositing in Nuke. Inventory the source media and confirm resolution, frame rate, color space, audio requirements, and delivery target. Preserve source files, create a new project, sequence, or comp, discover typed tools, complete one representative shot or sequence, render a preview, check missing media and frame range, and report the project and preview paths. Ask before installing codecs or plugins, relinking media, or starting a final render.

How do I create visual effects with AI?

Use Houdini for procedural geometry and simulation, Nuke or After Effects for shot compositing, and Unreal Engine for real-time effects. A good agent first identifies where the effect belongs, then builds and validates the smallest representative shot.

Use the dcc-mcp Skill to help me create a visual effect. First ask for the shot brief, reference, duration, camera, delivery format, and whether the effect belongs in Houdini simulation, Nuke or After Effects compositing, or Unreal real-time VFX. Inventory connected hosts, discover typed tools, and propose the smallest representative setup. Keep controls editable, preserve source plates and scenes, cache or render only when needed, produce a preview, validate timing, scale, color space, and missing dependencies, and report the scene, cache, and preview paths. Ask before downloading assets, installing plugins, or starting an expensive simulation or final render.

Why use DCC-MCP instead of asking for a script?

Typed discovery, validated arguments, instance routing, audit evidence, diagnostics, and consent boundaries are reusable across agents. The agent can change; the production contract stays stable. See For Agents for the complete workflow and failure-reporting rules.