These are the official MCP servers for images and design we found on 2026-11-06 — each published by the service itself, each checked on its own documentation. We make one of them (a template image API), so it is in the list, in its group, with the same facts as the others. Community wrappers are left out: anyone can wrap an API, and a wrapper's quality and permissions are the wrapper author's, not the service's.
How to read it: remote servers are reached at an URL (chat apps such as Claude and ChatGPT can use them); local servers run on your computer, started by a coding tool. OAuth means you sign in to the service in the browser; API key means you paste a key into the host's config.
Design tools
Figma. "The Figma MCP server brings Figma into your AI workflow. It gives AI agents the design context they need to generate code from your Figma files, and lets agents write native Figma content back to the canvas" (Figma). Remote at https://mcp.figma.com/mcp, sign in with Figma; a local desktop server needs a Dev or Full seat on a paid plan. Writing to the canvas needs the remote server and a Full seat, and "is free during the beta period and will eventually become a usage-based paid feature". Only clients in Figma's MCP catalog (VS Code, Cursor, Claude Code and others) can connect. Best for: design-to-code and editing real Figma files.
Canva. Two servers. The Canva MCP server (the "AI Connector") is remote and "gives your AI assistant access to Canva's design APIs and capabilities" — creating and editing designs, brand assets, export, comments — with OAuth sign-in (Canva Developers); Canva says it works on every account, Free included, with some actions such as Brand Templates needing a paid plan. The Canva Dev MCP server is local and helps developers build Canva apps. Best for: teams whose designs already live in Canva.
Adobe. The Adobe Express Developer MCP Server is local and helps developers build Express add-ons (Adobe Developer). Firefly Run-Workflow MCP is remote and lets Claude, Cursor or other clients "execute creative workflows using natural language", with an access token and API key from your organisation's admin — an enterprise feature (Adobe Help). Adobe's separate "Adobe connectors" for ChatGPT and Claude are not described as MCP servers on Adobe's page, so we leave them out.
Media libraries
Cloudinary. Five official servers — Asset Management, Environment Config, Structured Metadata, Analysis and MediaFlows — to "upload, manage, transform, generate, and analyze your media assets" (Cloudinary docs, updated 2026-11-05). Remote endpoints use OAuth by default, API keys in headers also work (not for Analysis); local npm packages exist too. Asset Management can generate images from a prompt with an add-on. Best for: a media library you already keep in Cloudinary — upload, crop, transform, tag.
AI image generation
Replicate. "The official hosted MCP server for Replicate", giving apps "access to Replicate's HTTP API so you can discover, compare, and run thousands of AI models" (Replicate). Remote with a Replicate API token, or local. Best for: trying and running open image models — Flux, SDXL and the like — from a chat.
fal. "The fal MCP server lets your assistant find models, check schemas and pricing, upload inputs, and run generations" (fal docs). Remote at https://mcp.fal.ai/mcp-relay, Streamable HTTP with OAuth, no API key needed. Best for: fast generative image and video models.
Generative models make new pictures; they do not reliably put your exact headline, logo and brand colours in the same place twice. That is what the next group is for.
Template image APIs
These render a fixed design with new text and images — the same layout every time, filled from data — which is what OG images, social cards, certificates and banners need.
Shareglint (ours). Remote at https://shareglint.com/mcp: OAuth for Claude and ChatGPT, an API key for coding tools. Tools: render_image, get_image, list_templates, get_template, create_signed_url, get_usage, generate_template, take_screenshot, auto_og_url, create_batch, get_batch — renders from 153 built-in templates or your own HTML and CSS, in 6 size presets, as PNG, JPEG, WebP, PDF, GIF or MP4, plus website screenshots and signed og:image URLs. Free plan 100 renders a month with a watermark (setup). Best for: developers who want HTML templates, screenshots and OG images from one account.
Bannerbear. "Generate images, videos and PDFs straight from Claude, Cursor or any MCP-enabled agent, using your Bannerbear templates" (Bannerbear). Hosted with OAuth ("no API key required") or local with an API key. Best for: teams already designing templates in Bannerbear's editor. Our Bannerbear comparison covers the differences.
Placid. "With our MCP, you can use Placid like a plugin for your AI applications", and "all outputs are still fully template-driven, and the MCP does not invent visuals" (Placid). A remote per-project URL with a bearer API token. See Placid vs us.
Templated. Remote at https://mcp.templated.io/mcp (listed in Claude's connector directory, sign in to Templated) or a local npm server with an API key (Templated). See Templated vs us.
Orshot. "Connect Orshot to Claude, Cursor, VS Code, ChatGPT and any MCP client to design templates, render and publish by chatting"; remote with OAuth, "no API keys needed" (Orshot).
APITemplate.io has no official MCP server that we could find on its site; the servers listed for it elsewhere are third-party wrappers (Zapier, Pipedream and others).
Diagrams and whiteboards
Excalidraw. An official server that "streams hand-drawn Excalidraw diagrams with smooth viewport camera control and interactive fullscreen editing", remote at https://mcp.excalidraw.com or local (GitHub); Excalidraw+ has its own MCP in public beta, with an API key.
tldraw. An MCP App launched in Cursor in March 2026 to let agents "draw, diagram, and visually collaborate", with other clients announced (tldraw blog).
Screenshots of web pages
Playwright MCP (Microsoft) gives agents "browser automation capabilities using Playwright" and includes a screenshot tool; it runs locally (npx @playwright/mcp@latest) (GitHub). Best for: agents that need to click through and test pages on your machine, localhost included. For screenshots of public pages from a chat app, without running a browser yourself, our server's take_screenshot does it remotely.
Choosing
- "Make me a picture of…" → Replicate or fal.
- "Change this design / turn this frame into code" → Figma or Canva.
- "Put this title on our OG card, for all 200 posts" → a template image API: ours, Bannerbear, Placid, Templated or Orshot.
- "Organise and transform our media library" → Cloudinary.
- "Draw an architecture diagram" → Excalidraw or tldraw.
- "Screenshot my local app while testing" → Playwright MCP; a public page from Claude or ChatGPT → a remote screenshot tool such as ours.
Whatever you pick, add servers from their publisher's own documentation, check what each tool can change on your account, and keep approval prompts on for actions that publish or delete. More on how servers work in what is an MCP server.
