Top 9 MCP Servers for Design & Media in 2026
Design and media are where AI agents move from "chat" to "make." A model that can read a Figma frame, generate a Flux image, dub a voiceover, or arrange a 3D scene in Blender turns a prompt into a finished asset — and the Model Context Protocol (MCP) is the plumbing that makes it possible. For anyone building creative tooling on top of Claude, Cursor, or Windsurf, the design and media category is one of the most active corners of the MCP ecosystem.
But there's a catch that most "top MCP servers" lists ignore: these servers run with real permissions on your machine. They open sockets to Blender and Ableton, drive Photoshop through a COM bridge, upload assets to your Cloudinary account, and hit third-party generation APIs with your keys. Star counts tell you what's popular — they tell you nothing about what the code actually does.
That's why every server below has been scanned by MCP Trust Checker, our deterministic security engine. Alongside each pick you'll see its MCP Trust Score: a letter grade (A–F) and a 0–100 number, plus any high-severity findings. Read it like a report card — A means the scan found nothing alarming; C or lower, or any high-severity finding, means you should review the details on the server's registry page before you grant it access. Here are the nine we'd actually recommend, best first.
1. Framelink Figma MCP
Framelink (originally Figma-Context-MCP) is the de-facto standard for design-to-code. Point it at a Figma file or frame URL and it returns a trimmed, agent-friendly version of the node tree — layout, styles, and component data — so a coding agent can implement the design in one pass instead of squinting at a screenshot. With around 14k GitHub stars it's the most-downloaded community Figma server and tops nearly every design MCP roundup.
Our scan gives it a clean bill of health with a minimal capability footprint, which is exactly what you want from a tool you'll wire into your daily editor.
MCP Trust Score: A (100/100) · GitHub · npm
2. Blender MCP
Blender MCP is the viral one — roughly 22k GitHub stars and probably the most popular creative MCP server overall. It connects a live Blender session to an LLM through an add-on plus MCP server, so an agent can build and edit 3D scenes, manipulate objects and materials, set up lighting, and even run Python inside Blender. It can pull assets from PolyHaven and Hyper3D Rodin, turning a text prompt into real modeling and rendering.
Honest note: our scan grades it C and flags 1 high-severity finding — unsurprising for a server that executes Python inside your Blender instance. It's genuinely useful, but review the finding on its registry page before you let an agent run code through it.
MCP Trust Score: C (76/100) · GitHub · PyPI
3. ElevenLabs MCP
This is the official server from ElevenLabs, and it's the default choice for audio in the MCP world. It exposes their text-to-speech, voice cloning, dubbing, sound-effect, and speech-to-text APIs as tools, so an agent can generate narration, clone a voice, transcribe an interview, or drop in sound effects straight from a prompt. If you're building any audio or video pipeline, this is a core building block.
Being a vendor-maintained package, it scans cleanly with no high-severity findings.
MCP Trust Score: A (98/100) · GitHub · PyPI
4. Cloudinary Asset Management MCP
Generation gets the headlines, but most real media workflows also need somewhere to store, transform, and deliver the results. Cloudinary's official server exposes its Upload and Admin APIs as tools, letting an agent upload, search, tag, transform, and organize images and video in your Cloudinary account. It handles the digital-asset-management backend rather than generation, and it's part of a suite of official Cloudinary MCP servers.
A perfect score in our scan, which is what you'd expect from a first-party package with a tight, well-defined scope.
MCP Trust Score: A (100/100) · GitHub · npm
5. Replicate Flux MCP
Replicate Flux MCP is one of the most-referenced community servers for prompt-to-image work. It generates high-quality images through Black Forest Labs' Flux Schnell model and vector graphics via Recraft V3 SVG, all over the Replicate API, with customizable generation parameters. It's a popular drop-in whenever you need an agent to produce raster or SVG assets on demand, and it works across MCP clients.
Clean scan, minimal capabilities — a solid default for image generation.
MCP Trust Score: A (100/100) · GitHub · npm
6. Photoshop MCP Server
For raster editing, this server drives Adobe Photoshop programmatically through its Python/COM API. An agent can create and save documents, add and manipulate layers (text, solid color), read session and document info, and apply effects and adjustments — great for automating the repetitive parts of a design job. It's actively maintained with 2026 releases and is the most-cited Photoshop automation server. One practical constraint: it's Windows-only, since it relies on the Photoshop COM bridge.
Our scan returns a clean grade with no high-severity findings.
MCP Trust Score: A (98/100) · GitHub · PyPI
7. Image Generation MCP Server (GongRzhe)
If Replicate Flux MCP feels like more than you need, GongRzhe's Image Generation server is the minimal alternative. It generates images from text prompts using Replicate's Flux model, with control over dimensions and output, and nothing else to get in the way. That simplicity makes it very easy to wire into Claude Desktop or Cursor, which is why it shows up so often on directories like mcp.so and mcpservers.org as a go-to entry-level image-gen option.
Perfect score in our scan.
MCP Trust Score: A (100/100) · GitHub · npm
8. Ableton Live MCP
From the same author as Blender MCP, this server bridges an LLM to a running Ableton Live session over a socket using a Remote Script. An agent can create tracks, edit MIDI clips, load instruments and effects, browse the library, and control playback — effectively composing and arranging music through conversation. With 2k+ stars it's the go-to for AI-driven music production and sound design.
As with Blender, the live-session bridge comes at a cost: our scan grades it C with 1 high-severity finding. Review it on the registry page before granting access, especially in a setup where the agent can execute arbitrary control commands.
MCP Trust Score: C (76/100) · GitHub · PyPI
9. EverArt MCP
EverArt MCP was one of the original showcase media servers for MCP — the reference image-generation server from the Model Context Protocol project itself, wrapping EverArt's API to produce 1024×1024 images across models like FLUX1.1, SD3.5, and Recraft. It's historically important and still widely cited. The important caveat: it now lives in the servers-archived repository and is deprecated, so treat it as a historical reference rather than something to build on today. It scans cleanly, but "clean and unmaintained" still means unmaintained.
MCP Trust Score: A (98/100) · GitHub · npm
How we graded these
Every Trust Score above comes from the MCP Trust Checker engine, which is deterministic and open-source. It reads the actual published source of each package from npm or PyPI — not a README, not a marketing page — and analyzes its capabilities and toxic-flow risk against a fixed rule set. There's no LLM in the loop and no judgment call: the same input always produces the same A–F grade and 0–100 score, so the numbers are reproducible and auditable. You can dig into the specific findings, capability level, and evidence for any server on its page in the Trust Registry.
This is just the shortlist. Browse every scanned server in this space at the full Design & Media category, and if a package you're eyeing isn't listed yet, scan it yourself for free through our public API — same deterministic engine, same score, in seconds.
Frequently asked questions
What is the best MCP server for design-to-code?
Framelink Figma MCP (the figma-developer-mcp package) is the community standard. It returns a trimmed, agent-friendly version of a Figma file or frame so a coding agent can implement the design directly, and it scans A (100/100) with no high-severity findings in the MCP Trust Registry.
Are design and media MCP servers safe to use?
Most are, but it varies. Vendor-maintained servers like ElevenLabs and Cloudinary scan cleanly, while servers that bridge to a live app and run code — such as Blender MCP and Ableton Live MCP — grade C with one high-severity finding each. Always check a server's MCP Trust Score and review its findings on the registry page before granting it access to your machine or accounts.
Which MCP server should I use for AI image generation?
For flexible prompt-to-image and prompt-to-SVG work, Replicate Flux MCP (A, 100/100) is a widely used choice. If you want the simplest possible setup for Claude Desktop or Cursor, GongRzhe's Image Generation MCP Server (A, 100/100) is a minimal Flux-based option. Both scan cleanly in our registry.
What does the MCP Trust Score mean?
It's a deterministic A–F grade and 0–100 score produced by MCP Trust Checker, which reads a package's actual published source from npm or PyPI and analyzes its capabilities and toxic-flow risk against a fixed rule set. No LLM is involved, so the same input always yields the same score. An A means nothing alarming was found; a lower grade or any high-severity finding is a signal to review the details before using the server.
Is there a good MCP server for 3D and audio work?
Yes. Blender MCP connects a live Blender session to an LLM for prompt-driven 3D modeling and rendering, and Ableton Live MCP does the same for music production in Ableton. Both are popular and capable but grade C with one high-severity finding in our scan, so review their registry pages before use. For voice and audio generation specifically, the official ElevenLabs MCP scans A (98/100).
Scan your MCP server now
MCP Trust Checker is free, open-source and runs entirely on your machine. Get an A–F Trust Score for any MCP server in seconds.
npx mcptrustchecker