Best Design & Creative MCP Servers in 2026

Design and creative MCP servers connect an AI agent to the tools that make visual deliverables — UI designs, 3D scenes, diagrams, slides, videos, voiceovers, images, and product mockups. For installable MCP servers an agent runs directly, the two top picks are Figma (read design context from real files and generate spec-accurate UI code) and Blender (drive a live 3D session — create objects, set materials, run Python, render). Alongside them, a wide set of in-ChatGPT creative apps cover decks (Gamma, Beautiful.ai, SlidesGPT), diagrams (Lucid, Whimsical), video (HeyGen, invideo, VEED), stock and assets (Cloudinary, Shutterstock), voice (Speechify, AI Voice Generator), and AI app-builders (Lovable, Base44, Adalo, Manus, B12) that ship working frontends straight from prompts.

Choose by deliverable. For "turn this Figma design into code," Figma is the default — it pulls frames, components, and design tokens so the agent stops inventing CSS. For "model or edit a 3D scene," Blender is the only credibly-top MCP server, with viewport screenshots so the agent can see and correct its own work. For decks start with Gamma; for diagrams Lucid and Whimsical; for short-form video invideo and VEED; for "build me a working app from this brief," the AI app-builders are remarkable but still need a developer in the loop for production code. Pair any creative MCP with a Filesystem MCP so the agent can save outputs back to disk.

Common mistakes: letting Blender run agent-written Python against an unsaved .blend file (version-control it first — the execution is not sandboxed), accepting AI-generated images without a usage-rights check, and forgetting that generative video and voice cost real money per minute — set quotas. Every entry below documents which formats it handles natively, licensing posture, auth, and per-call pricing where relevant. Start with one tool per deliverable type, prove a single end-to-end output, then layer in alternatives only when the gap is concrete.