DCC stands for Digital Content Creation — the 3D, CAD, GIS and imaging tools studios already run, like Blender, Maya, Houdini and Photoshop.
MCP for Blender, Rhino, QGIS & Engineering Software
Connect any AI agent to your creative and engineering tools — securely and reliably. Deploy production-ready MCP Servers and plugins that let Claude, Cursor and your own agents inspect, build and validate work inside Blender, Maya, Houdini, Rhino, Photoshop, FreeCAD, QGIS and more, in real time.
Works with Claude Code, Codex, Cursor, OpenClaw, Doubao, Qwen and any other MCP-compatible agent.
- Read-only by default
- Checkpointed write tools
- stdiolocal transport
- 10
- host applications supported
- 90+
- typed tools, each risk-classed
- Local
- stdio first, nothing leaves your machine
Example session: an AI agent driving Blender through DCCMCP tools. Build three facade variants with 20 / 40 / 60 mm window insets.. scene.inspect({ depth: "collections" }). 42 objects · 6 collections. geometry_nodes.set_input({ socket: "Inset", value: 0.02 }). socket type Length validated. render.capture({ samples: 64 }). 3 variants rendered
Recorded session format · tool calls shown are real DCCMCP tool names
Why most AI-to-DCC integrations fail in production
The model writes a script it has never run
Ask an agent for bpy or PyQGIS code and it guesses. Object names, collection paths, units, tolerances — none of it is verified against your file. You find out at the traceback, halfway through a change.
One bad call costs an afternoon
A destructive operator without a checkpoint can corrupt a scene, break linked libraries or invalidate a parametric tree. There is no undo for the two hours before it.
Nobody can prove what changed
Unattended automation with no record of who changed what, when and with which arguments cannot pass studio sign-off, client review or a manufacturing audit.
The hidden cost of “just let the model write a Python script” is not the generation. It is the debugging, the review burden, the corrupted-file incidents and the audit trail that does not exist when anyone asks what happened on Thursday.
A standard contract between your agent and your software
Install the MCP server
Add the integration for your host application — Blender addon, Rhino Yak package, FreeCAD workbench, QGIS plugin or a Python package for OpenCV.
npx dccmcp-blender@latest serveConnect your agent
Register the server with Claude Code, Codex, Cursor, OpenClaw, VS Code or your own MCP client. Local stdio by default; streamable HTTP when a studio server needs to serve a team.
"{ "mcpServers": { "blender": … } }"Control the software
The agent inspects real state and calls typed tools. Invalid calls are rejected at the boundary; valid ones edit geometry, run algorithms and export deliverables.
geometry_nodes.set_input({ socket: "Inset" })Verify and audit
Every call is logged with its arguments and result. Checkpoints make destructive work reversible, and approvals gate anything that touches production files.
dccmcp audit tail --since 2hMCP Servers for the software you already run
10 integrations · 90+ typed tools
MCP for 3ds Max, MCP for ZBrush and MCP for OpenCV are documented on their own pages, along with the full tool list and support matrix for every integration. If your software is not listed, we build private bridges on the enterprise track — Revit, Civil 3D, KiCad, Ansys, Maya-based in-house tooling and more.
Works with the agent you already use
Not on the list? If it speaks MCP, it works
Add the server exactly the way you add any other MCP server: one config entry, stdio by default, streamable HTTP with token auth for a shared studio deployment. Custom agents built on the MCP SDK get the same typed tools.
MCP support is implemented by each client, not by us. Where a client lets you add an MCP server, add the DCCMCP entry and the tools appear — the same config works in every client listed above.
Supported agent clients include Claude Code, Codex, Cursor, GitHub Copilot, Claude Desktop, Gemini CLI, Trae, Doubao, Qwen Code, CodeBuddy, Custom MCP clients.
Built for pipelines that cannot be interrupted
Production-grade safety
Read-only by default. Approval gates on write tools, deny rules for production paths, and per-tool permissions that a TD can configure once and enforce everywhere.
Auditable by default
Every call records the tool, arguments, result, host version and file fingerprint. Export to your SIEM or query it locally.
Multi-instance control
Drive several sessions of the same application, or a mixed set of applications, from one agent run. Headless workers included.
Skills & custom tools
Register your own typed tools and packaged skills so agents can use studio conventions instead of inventing them.
Real UI control
Tools act on the live document, not a copy. The viewport, layer tree and undo stack stay in sync while the agent works.
Fast enough to stay in flow
Tool dispatch is a local call, not a network round-trip. Results stream back to the agent as they land, so long operations report progress instead of stalling.
- 4
- transports
- 0
- cloud deps
- 24/7
- local
Four stages, every session
Scroll through a single agent session. Each stage maps to a guarantee in the tool contract, so the behavior is the same whether the agent is a chatbot, a CI job or a pipeline you built in-house.
{
"file": "facade_module_01.blend",
"units": { "system": "metric", "scale": 1.0 },
"collections": ["Facade_Module", "Site", "Render_Setup"],
"objects": 42,
"checkpoint": "auto/facade-a1.ckpt"
}We would rather show you the contract than a logo wall
Published tool surfaces
Every integration documents its tools, arguments and risk class in public. You can read exactly what an agent is able to do before you install anything.
A written security posture
Local-first transport, read-only defaults, checkpointed writes and exportable audit logs — described as a policy you can hand to your IT team.
A dated support matrix
We publish which host application versions each release supports, and how long after a vendor release we ship an update. No silent breakage.
Design partner program
We are onboarding studios, AEC firms and engineering teams now. Design partners get Studio seats at no cost during the program, direct access to the engineers building the integrations, and a vote on what ships next.
Free at launch, upgrade when a team depends on it
Community
Solo artists, students and evaluation
Everything a solo designer or engineer needs to connect an agent to one machine. Free at launch — no account, no card, no call.
- All public integrations (Blender, Rhino, FreeCAD, QGIS, OpenCV)
- Read-only-by-default safety policy
- Local stdio transport
- Community support and Discord
- Unlimited local sessions
Studio
Freelancers and small studios shipping client work
Production tooling for working studios: multi-instance control, audit logs and shared policies.
- Everything in Community
- Multi-instance and headless/batch control
- Audit logs with 90-day retention
- Team policy profiles and approval gates
- Custom tool registration (Skills SDK)
Enterprise
Studios, AEC firms and manufacturers with IT requirements
For pipelines that cannot be interrupted: SSO, self-hosting, compliance reviews and custom integrations.
- Everything in Studio
- SSO/SAML, SCIM and role-based access control
- Self-hosted or air-gapped deployment
- Unlimited audit retention with SIEM export
- Custom integration development
Questions teams ask before they install
The Model Context Protocol is a standard way for AI agents to discover and call tools. Applied to creative and engineering software, it replaces one-off generated scripts with a stable, reviewable interface — so an agent works against a contract instead of guessing at an API.
Any MCP-compatible client. We verify Claude Code, Cursor, Codex, GitHub Copilot, Claude Desktop, Gemini CLI, Windsurf, Cline, Roo Code, Trae, Doubao, Qwen Code, Kimi, Zhipu GLM, CodeBuddy, Yuanbao, Lingma, Qoder, Continue, Zed, JetBrains AI, Goose, Warp, Coze, CodeGeeX, Comate, ERNIE, Spark, MiniMax Agent, OpenClaw and WorkBuddy against each release. Agents you build yourself on the MCP SDK get the same typed tools, over stdio or streamable HTTP.
Not on Community or Studio. The server runs locally next to your application and only the tool results an agent requested are passed to your model provider. Enterprise deployments can run fully self-hosted or air-gapped.
Blender 4.2 LTS and newer, Maya 2024+, Houdini 20.5+, 3ds Max 2024+, Rhino 7 and 8, ZBrush 2024+, Photoshop 2024+, FreeCAD 0.21 and 1.0+, QGIS 3.34 LTR and newer, and OpenCV 4.x/5.x. Each integration page lists the exact support matrix, and legacy versions are available on the enterprise track.
Yes. The Skills SDK lets you register your own typed tools against any Python or .NET surface, and enterprise customers can commission fully maintained private bridges.
We focus on the operational layer that production pipelines need: policy enforcement, approval gates, checkpointed writes, audit trails, multi-instance control and vendor-version support. Where open-source components are useful, we use and credit them — with a clear separation between our commercial product and any open-source dependency.
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