Cloudinary MCP Server

The Cloudinary MCP servers connect AI assistants such as Claude, Cursor, Windsurf, or Visual Studio Code to Cloudinary, a platform for managing, transforming, and delivering images and videos. Instead of searching for media in a web portal, tagging by hand, or hunting for settings across several menus, you state your request in conversation: the assistant uploads files, finds assets, creates folders, and changes configurations. The servers are operated by Cloudinary itself. This is therefore not a community solution but an official offering from the platform operator.

What are the Cloudinary MCP servers?

Cloudinary does not offer a single server but a family with clearly separated tasks. According to the repository cloudinary/mcp-servers and the official documentation there are five servers. The Asset Management server handles uploading, managing, searching, folders, tags, and transformations. The Environment Config server manages upload presets, upload mappings, named transformations, webhooks, and streaming profiles. The Structured Metadata server creates metadata fields, values, and conditional rules. The Analysis server provides automatic tagging, moderation, and object detection. In addition there is a MediaFlows server for low-code automations around images and videos. The split helps you give an assistant only the permissions it needs for its task.

Remote or local

The recommended option is the remote servers hosted by Cloudinary, for example https://asset-management.mcp.cloudinary.com/mcp. They need no installation. For new configurations the repository recommends the /mcp endpoint with Streamable HTTP; the older /sse endpoint is kept for backward compatibility. If you want more control, you start the servers locally through npx from npm packages, which requires Node.js 18 or later, and set the credentials in environment variables. According to the repository there are also official Docker images on Docker Hub.

What are they particularly useful for?

The value lies in recurring maintenance work. Anyone looking after a large media library can ask the assistant to find images with a certain tag, move them to a new folder, or create variants with other dimensions. Developers have upload presets and webhooks created in the editor without opening the console. Editorial teams use structured metadata to set up mandatory fields and rules consistently. Compared with a generic file server, this server knows Cloudinary's concepts and limits, such as transformations, derived assets, or regions, and therefore returns more fitting results.

Setup and sign-in

By default the remote servers use OAuth; on first connection the client opens the browser. Alternatively, API credentials can be passed as headers, either bundled as cloudinary-url or individually for cloud name, API key, and API secret. According to the documentation the Analysis server supports OAuth only. Local servers use the variables CLOUDINARY_URL or CLOUDINARY_CLOUD_NAME, CLOUDINARY_API_KEY, and CLOUDINARY_API_SECRET. Headers can also set the region and a selection of the exposed tools, which reduces the attack surface. For Claude and Cursor there are also marketplace plugins that bundle skills and the remote servers.

Security and limits

An assistant with write permissions can overwrite files, change presets, or redirect webhooks. Start with a test environment, select only the tools you need, and never store the API secret in the repository. The documentation does not list all tools; check in your client which actions are actually offered. Content the assistant retrieves enters the AI client context and therefore reaches the model provider. According to the provider, some features, such as automatic tagging or image generation, require an activated add-on.

Costs

The servers themselves are open source under the MIT license but require a Cloudinary account. According to the pricing page there is a permanently free plan for the image and video APIs, paid plans such as Plus and Advanced, and an Enterprise tier by individual agreement; for digital asset management there is a time-limited trial. Which terms apply to you is shown on Cloudinary's official pricing page.

Who are the Cloudinary MCP servers for?

They are worthwhile for teams that already work with Cloudinary and want to handle maintenance, configuration, and review in conversation: web developers, editorial teams, e-commerce teams with many product images, and agencies with several client environments. Anyone without a Cloudinary account gains nothing from the servers, since they are an access path to the platform, not a standalone image editor. For single files on your own computer, other tools are simpler.

A Cloudinary account (the free plan is enough to try it out) and an MCP client that supports remote servers, such as Claude Code or Cursor. Local servers additionally need Node.js 18 or later.

With the Cloudinary MCP servers, an AI assistant manages your images and videos in conversation. This guide leads from sign-in to the first task.

Step 1: Check the prerequisites

You need a Cloudinary account and an MCP client that supports remote servers, such as Claude Code or Cursor. According to the pricing page the free plan is enough to try it out. It is best to begin in a test environment.

Step 2: Register the server

The Asset Management server has the address https://asset-management.mcp.cloudinary.com/mcp. In Claude Code you register it with claude mcp add --transport http cloudinary-asset-management https://asset-management.mcp.cloudinary.com/mcp. Other servers such as Environment Config or Structured Metadata are registered with the respective address from the documentation.

claude mcp add --transport http cloudinary-asset-management https://asset-management.mcp.cloudinary.com/mcp

Step 3: Sign in

On first connection the client opens the browser and you confirm access via OAuth. Alternatively you can set the header cloudinary-url with your API credentials; keep them in an environment variable, never in the repository. The Analysis server supports OAuth only.

Limit tools

Use the header cloudinary-tools to select the exposed tools and cloudinary-region for the region of your environment. Give the assistant only what the task needs, especially if it could overwrite or delete files.

Step 4: First tasks

Start with read questions such as "Which folders exist?" or "Show images with the tag summer". Then have files uploaded, tags set, or assets moved. Check the results in the Cloudinary interface.

Step 5: Running locally

If you want to run the servers locally, start them with npx -y --package @cloudinary/asset-management-mcp -- mcp start and set CLOUDINARY_URL as an environment variable. This requires Node.js 18 or later.

npx -y --package @cloudinary/asset-management-mcp -- mcp start

OAuth via the browser with your own Cloudinary account; alternatively API credentials as a header (cloudinary-url) or, for local servers, through environment variables.

Read and write access to media and configuration of the signed-in Cloudinary environment; the exposed tools can be restricted through a header or start option.

The remote servers run hosted by Cloudinary. Retrieved content and metadata enter the AI client context and therefore reach the model provider.

Unwanted deletion or overwriting of media, changed upload presets or webhooks, passing media content to the model provider, and an API secret stored in the repository.