Provides a Unix-style shell, scriptable CLI, and MCP server to manage Cloudinary assets like a local file system.
Also known as Sheldon1999/cldsh-Cloudinary-Shell, cloudinary-shell
- Stars
- 0
- Tools
- 11
- Downloads, 30 days
- —
- Latest version
- 2.0.1
- Last commit
- Oct 28, 2026
- License
- MIT
What is Cloudinary Shell MCP?
Cloudinary Shell provides a Unix-style shell, scriptable CLI, and MCP server to manage Cloudinary assets like a local file system. It exposes 11 tools, such as cloudinary_list, cloudinary_tree, cloudinary_info.
Summary generated from the project's README
About Cloudinary Shell MCP server
Cloudinary Shell provides a unified interface for managing Cloudinary assets through a Unix-style shell, a scriptable CLI, and an MCP server. It allows developers to interact with their Cloudinary account as if it were a local file system, supporting both dynamic and fixed folder modes. The tool is built as a single Node.js binary that requires an API key or a Cloudinary URL for authentication, eliminating the need for Python runtimes or OAuth browser flows.
AI agents using Claude Desktop, Claude Code, or Cursor can utilize the MCP server to perform tasks like listing assets, uploading files, or generating delivery URLs. The server includes built-in safety mechanisms, such as a two-step confirmation process for destructive actions like deleting or moving assets. For automation, the CLI offers machine-readable JSON output and specific exit codes to facilitate integration into CI/CD pipelines and custom scripts.
What you can do with it
- 1Automate asset management in CI/CDA developer can use the CLI to upload a build directory to a specific Cloudinary folder using a single command.
- 2Interactive asset browsing and previewingDevelopers can use the interactive shell to navigate folders, use tab completion, and view image or video thumbnails directly in the terminal.
- 3Extracting asset metadata for scriptsYou can pipe JSON output from the tool into other utilities to filter assets based on specific criteria like file size.
Written from the project's README. Every use case is checked against a quote from it.
How to install Cloudinary Shell MCP
Formats verified 10 Nov 2026Install Cloudinary Shell MCP in Claude Code
$ claude mcp add cldsh-cloudinary-shell -- npx -y cloudinary-shellInstall Cloudinary Shell MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"cldsh-cloudinary-shell": {
"command": "npx",
"args": [
"-y",
"cloudinary-shell"
]
}
}
}Install Cloudinary Shell MCP in VS Code
.vscode/mcp.json
{
"servers": {
"cldsh-cloudinary-shell": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"cloudinary-shell"
]
}
}
}Install Cloudinary Shell MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"cldsh-cloudinary-shell": {
"command": "npx",
"args": [
"-y",
"cloudinary-shell"
]
}
}
}Install Cloudinary Shell MCP in Codex
~/.codex/config.toml
[mcp_servers.cldsh-cloudinary-shell] command = "npx" args = ["-y", "cloudinary-shell"]
Install Cloudinary Shell MCP in Gemini CLI
~/.gemini/settings.json
{
"mcpServers": {
"cldsh-cloudinary-shell": {
"command": "npx",
"args": [
"-y",
"cloudinary-shell"
]
}
}
}Cloudinary Shell MCP tools 11
Detected from the READMEcloudinary_listcloudinary_treecloudinary_infocloudinary_get_urlcloudinary_uploadcloudinary_downloadcloudinary_movecloudinary_deletecloudinary_mkdircloudinary_rename_foldercloudinary_account
| cloudinary_list | — | |
| cloudinary_tree | — | |
| cloudinary_info | — | |
| cloudinary_get_url | — | |
| cloudinary_upload | — | |
| cloudinary_download | — | |
| cloudinary_move | — | |
| cloudinary_delete | — | |
| cloudinary_mkdir | — | |
| cloudinary_rename_folder | — | |
| cloudinary_account | — |
Automated checks
Ran Nov 14, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work
- Source repository linked and reachablesheldon1999/cldsh-cloudinary-shell is public and reachablegithub.com/sheldon1999/cldsh-cloudinary-shell ↗Pass
- Package names this repository as its sourcecloudinary-shell names sheldon1999/cldsh-cloudinary-shell as sourcenpmjs.com/package/cloudinary-shell ↗Pass
- Build provenance (Sigstore attestation)No build attestation published for this releaseNot checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-10-28github.com/sheldon1999/cldsh-cloudinary-shell ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNo popular package one edit away from cloudinary-shellPass
- No critical advisory in the package or its dependencies4 resolved dependencies checked; no critical advisoryosv.dev ↗Pass
- No hidden instructions to the model11 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNo tool named like delete, drop or execPass
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubCloudinary Shell (cldsh)
MCP server, interactive shell and scriptable CLI for Cloudinary

cldsh lets you work with a Cloudinary account like a file system:
- AI agents (Claude Desktop, Claude Code, Cursor) get Cloudinary tools through the Model Context Protocol (MCP), with previews and confirmation before anything is deleted or replaced.
- Developers get an interactive shell with folders, tab completion, history and inline image previews.
- Scripts and CI get single commands with JSON output and meaningful exit codes.
It works with both Cloudinary folder modes: dynamic folders (every account created since June 2024) and the older fixed folders. The mode is detected automatically.
Working on a project that uses Cloudinary? Give
INTEGRATION.mdto your AI agent or team: integration patterns (Next.js, React, Node.js, Python), transformation recipes andcldshworkflows.
How cldsh compares
Cloudinary also publishes an official CLI (cloudinary-cli, Python) and official MCP packages (@cloudinary/*). cldsh is a separate, independent project built for a different workflow: one Node binary that is a shell, a scriptable CLI and an MCP server together, authenticated with your existing API key — no Python runtime, no OAuth browser flow, no hosted service required.
cldsh |
cloudinary-cli (official) |
@cloudinary/* MCP (official) |
community MCP servers | |
|---|---|---|---|---|
| Runtime | Node, npx |
Python, pipx/Docker |
Node, npx per package |
Node, npx |
| Auth | API key / .env |
OAuth login | OAuth (hosted) or API key | API key |
Interactive shell (cd, ls, tab completion) |
✅ | ❌ one-shot flags | ❌ not a CLI | ❌ not a CLI |
| MCP server | ✅ same binary | ❌ | ✅ (5 separate packages) | ✅ |
| Confirm before delete/move/overwrite | ✅ two-step token, --dry-run |
❌ | — | ❌ |
| Fixed/dynamic folder mode | ✅ auto-detected | not documented | not documented | not documented |
Pick cldsh for one local tool, API-key auth and a safety net on AI-driven changes. Pick the official packages for Cloudinary-maintained, domain-split MCP servers with OAuth and enterprise support.
Quick start
Requires Node.js 20 or newer and a Cloudinary account (cloud name, API key and API secret, or a CLOUDINARY_URL).
# AI agents: add the MCP server to Claude Code
claude mcp add cloudinary npx -y cloudinary-shell --mcp
# Developers: open the interactive shell
npx cloudinary-shell
# Scripts: run one command
npx cloudinary-shell ls products --json
Install globally to get the shorter cldsh command: npm install -g cloudinary-shell.
MCP server (for AI agents)
cldsh --mcp runs an MCP server over stdio (JSON-RPC 2.0). It supports protocol versions 2025-06-18, 2025-03-26 and 2024-11-05.
Client setup
Claude Desktop (claude_desktop_config.json):
{
"mcpServers": {
"cloudinary": {
"command": "npx",
"args": ["-y", "cloudinary-shell", "--mcp"]
}
}
}
Claude Code: claude mcp add cloudinary npx -y cloudinary-shell --mcp
Cursor: Settings → MCP → Add new MCP server, type command, command npx -y cloudinary-shell --mcp.
Check the setup with cldsh mcp: it runs the handshake, lists the tools and makes a live call on your account.
Tools
| Tool | What it does | Main arguments |
|---|---|---|
cloudinary_list |
List folders and assets (wildcards allowed) | path, resource_type, delivery_type, max |
cloudinary_tree |
Folder tree | folder, depth, resource_type, max_files |
cloudinary_info |
Full metadata of one asset | public_id, resource_type, delivery_type |
cloudinary_get_url |
Delivery URL, optionally transformed or signed | public_id, transformation, signed, resource_type |
cloudinary_upload |
Upload a local file or directory | local_path, name, folder, recursive, overwrite, resource_type, dry_run |
cloudinary_download |
Download assets to the local disk | public_id, destination, overwrite, dry_run |
cloudinary_move |
Move or rename an asset or folder | source, target, overwrite, change_public_id, dry_run |
cloudinary_delete |
Delete assets, or folders with recursive |
target, recursive, resource_type, dry_run |
cloudinary_mkdir |
Create a folder | folder |
cloudinary_rename_folder |
Rename or move a folder (dynamic folder mode) | from, to |
cloudinary_account |
Cloud name, folder mode, credential source, remaining API budget | — |
Every tool carries MCP annotations (read-only, destructive, idempotent), so clients can decide when to ask you.
Safety
- Two-step changes.
cloudinary_delete,cloudinary_move,cloudinary_rename_folder, and uploads that would replace existing assets first return a preview with aconfirm_token. Nothing changes until the agent calls again with the same arguments and that token. Tokens are single-use, expire after 5 minutes and stop working if the account changes.dry_run: trueonly previews. - Local files. Uploads and downloads can only reach the folder the server was started in, the folders your MCP client shares as roots, and folders added with
--allow-path. The cldsh credentials folder,~/.sshand hidden files (unless named explicitly) are always refused. Downloads never overwrite a local file unlessoverwrite: true. - Server options:
--read-only: only the read-only tools (list, tree, info, get_url, account).--allow-path <dir>: allow another local folder (repeatable).--no-confirm: run changes in one call, as in cldsh 1.x. Use only for trusted automation.
"args": ["-y", "cloudinary-shell", "--mcp", "--read-only"]
Command line (for scripts and CI)
Run any command directly: cldsh <command> [arguments] [options].
- Results go to stdout; progress, warnings and errors go to stderr.
--jsonprints machine-readable JSON on stdout (errors are printed as JSON on stderr).- Exit codes:
0success,1error,2wrong usage,3partly failed (for example 2 of 50 uploads failed),4needs confirmation (add--yes) or was cancelled. --profile <name>uses a saved profile for one command.
# Upload a build folder with subfolders
cldsh upload ./dist/assets -r --to site/assets
# CDN URLs of every asset in a folder
cldsh ls portfolio --json | jq -r '.files[].secure_url'
# Assets larger than 100 KB
cldsh ls portfolio --json | jq -r '.files[] | select(.bytes > 100000) | .public_id'
# Preview a delete, then run it without the question
cldsh rm 'old-*' --dry-run
cldsh rm -r archive/2023 --yes
Quote wildcards. Your shell expands an unquoted
cld*into local file names beforecldshsees it. Writecldsh rm 'cld*'. If every name given tormexists in the local directory,cldshrefuses and reminds you.
Interactive shell
Run cldsh with no arguments.
Connected to Cloudinary account: my-cloud (global credentials profile (default))
[my-cloud]$~ cd products
[my-cloud]$~/products ls -l
SIZE(KB) DIMENSIONS(px) NAME FORMAT TYPE
──────── ────────────── ──────── ────── ─────
245 1200x630 hero png image
15 512x512 logo svg image
1830 1920x1080 promo mp4 video
- The prompt shows the account (and profile) and the current folder.
- Tab completes commands, options, folder paths (
cd apps/<Tab>) and asset names. - History is kept across sessions in
~/.cldsh_history(private file; lines with credentials are never saved). - Paths are relative to the current folder;
/or~is the root,..goes up,cd -goes back. helplists all commands;help <command>lists all options of one command.
Commands
| Command | What it does |
|---|---|
login [profile] [URL] [--stdin] |
Connect an account and save it as a profile (checked with Cloudinary first) |
config [ls | use <profile> | rm <profile>] |
Show the active account and which sources it overrides; list, switch or remove profiles |
ls [path|pattern] [-l] |
List folders and assets |
cd, pwd |
Change and show the current Cloudinary folder (shell only) |
lcd, lpwd |
Change and show the local directory |
mkdir <folder>... |
Create folders (parents included) |
upload <path> [name] [--to folder] [-r] |
Upload a file or a directory |
download <name|pattern>... [destination] [-f] |
Download assets |
mv <source> <target> [-o] [--public-id] |
Move or rename an asset or folder |
info <name> |
Asset metadata (-v for the full API response) |
url <name|pattern>... [--t=...] [--signed] |
Delivery URLs |
view <name> (alias preview) |
Show an image or video thumbnail in the terminal |
open <name> |
Open an asset in the browser |
rm [-r] <name|folder|pattern>... [-y] |
Delete assets, or folders with -r |
tree [folder] [-d N] |
Folder tree |
quota |
Remaining Admin API requests this hour and the folder mode |
mcp |
Check the MCP server |
Options shared by many commands:
| Option | Meaning |
|---|---|
--rt=image|video|raw|all |
Resource type. Listings show all types by default. |
--type=upload|private|authenticated |
Delivery type |
--dry-run |
Show what would happen without changing anything |
-y, --yes |
Don't ask for confirmation |
--json |
JSON output |
Assets can be named by their name in the current folder, by path (products/hero), by public ID, or as asset_id:<id>. If a name matches several assets (for example an image and a video called intro), cldsh lists them; add --rt or use the public ID.
Uploading
[my-cloud]$~/products upload ~/Pictures/hero.png
Uploaded (created): https://res.cloudinary.com/my-cloud/image/upload/v1730000000/products/hero.png
[my-cloud]$~ upload ./media -r --to campaigns/fall
[1/3] created https://res.cloudinary.com/my-cloud/image/upload/.../campaigns/fall/banner.png
[2/3] created https://res.cloudinary.com/my-cloud/video/upload/.../campaigns/fall/teaser.mp4
[3/3] created https://res.cloudinary.com/my-cloud/raw/upload/.../campaigns/fall/notes.txt
- The resource type comes from the file extension: images (including PDF and SVG) → image, video and audio → video, everything else → raw.
--rtoverrides it. - Images and videos get a public ID without extension; raw files keep it (
notes.txt). -rincludes subdirectories as subfolders. Dotfiles are skipped.- Existing assets are overwritten and reported as
overwritten;--no-overwritekeeps them (skipped). - When two files would get the same public ID (
photo.jpg,photo.png), both get their extension as suffix (photo_jpg,photo_png). - Files over 20 MB are uploaded in chunks with progress; temporary failures are retried.
--dry-runshows the planned public IDs and works without credentials.
Deleting
[my-cloud]$~/products rm hero
deleted products/hero
Deleted 1 · not found 0 · failed 0
[my-cloud]$~ rm -r campaigns/fall
Delete 3 assets (1 image, 1 video, 1 raw, 2.1 MB) and 1 folder in cloud my-cloud:
...
Proceed? [y/N]
rm -rdeletes everything in the folder and its subfolders (all resource and delivery types), then the folder.- Recursive deletes, wildcard deletes and deletes of more than 20 assets show the plan and ask first. Emptying the whole account asks you to type the cloud name.
-yskips the question; scripts need--yes. - Every asset is reported as
deleted,not foundorfailed, and the exit code is3if anything was not found or failed. --skip-backupdeletes folders that still hold backups of deleted assets.
Moving and renaming
[my-cloud]$~/products mv hero archive/
Moved: /products/hero -> /archive/hero (URL unchanged)
- Dynamic folder mode: moving and renaming keep the delivery URL (the folder and display name change).
--public-idalso renames the public ID, which changes the URL. Folders can be moved and renamed:mv products catalog/products. - Fixed folder mode: the folder is part of the public ID, so moving or renaming changes the URL. Cloudinary can't move folders in this mode.
- An existing target stops the move;
-ogoes ahead. In dynamic mode,-oonly replaces an asset with the same public ID; an asset that merely has the same name is kept. --dry-runshows whether the URL will change.
URLs, previews and downloads
url hero # delivery URL
url hero --t="f_auto,q_auto,w_800" # with a transformation
url 'banner-*' # several assets
url private-doc --signed # signed URL for authenticated assets
view hero --t="e_grayscale" # inline preview (iTerm2, WezTerm, Ghostty, VS Code, Warp)
download 'shots/*' ./shots/ # several files into a folder
- Unsigned URLs without
--tare exactly the asset'ssecure_url. viewprints the preview URL when the terminal can't show images.downloadskips local files that already exist unless-f; private and authenticated assets download through signed URLs.
Accounts and credentials
# Interactive (the secret is not shown while you type)
cldsh login
# From a script, without putting the secret in your shell history
printf %s "$CLOUDINARY_URL" | cldsh login prod --stdin
cldsh config ls # saved profiles
cldsh config use prod # switch (also inside the shell)
cldsh config # active account, where it comes from, what it overrides
Profiles are stored in ~/.config/cldsh/credentials.json (%APPDATA%\cldsh on Windows) with owner-only permissions. Writes are atomic, and a damaged file is never overwritten (login --reset moves it aside).
When several sources are present, the strongest wins:
--profile <name>,config use/loginin the current shell, orCLOUDINARY_PROFILECLOUDINARY_URL(orCLOUDINARY_CLOUD_NAME,CLOUDINARY_API_KEY,CLOUDINARY_API_SECRET) in the environment- A
.envfile in the current directory - The saved default profile
cldsh warns at startup, and config shows, when a source is overridden. CLOUDINARY_URL options such as ?upload_prefix=https://api-eu.cloudinary.com (EU/AP data centers) and private CDN settings are supported.
Folder modes and API limits
cldshdetects the folder mode of your account. Override it with--folder-mode=fixed|dynamicorCLDSH_FOLDER_MODE.- Listing folders and assets uses Cloudinary's Admin API, which is rate-limited per hour (500 requests on the Free plan).
cldshcaches listings for a few seconds, warns when less than 10% of the hourly budget is left, andquotashows the remaining requests. Uploads and delivery URLs don't count. - In fixed folder mode, listing a folder also pages through its subfolders' assets, so very large folders cost more requests.
Troubleshooting
| Problem | What to do |
|---|---|
Not connected to a Cloudinary account |
Run cldsh login, or set CLOUDINARY_URL. cldsh config shows what is active. |
| The wrong account is used | cldsh config lists the source in use and the ones it overrides. |
Refusing to run "rm" on ... names that all exist in the local directory |
Your shell expanded a wildcard; quote it: cldsh rm 'cld*'. |
needs an interactive terminal / exit code 4 |
The command wants a confirmation; add --yes in scripts. |
matches 2 assets |
Add --rt=image (or video/raw) or use the public ID. |
| Rate limit reached | cldsh quota shows when the hourly budget resets. |
License
MIT. See CHANGELOG.md for changes between versions.
Cloudinary Shell MCP questions
How do I add Cloudinary Shell MCP to Claude Code?
Run claude mcp add cldsh-cloudinary-shell -- npx -y cloudinary-shell in your terminal. Add --scope user to make it available in every project.
How do I add Cloudinary Shell MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
What tools does Cloudinary Shell MCP provide?
It exposes 11 tools, including cloudinary_list, cloudinary_tree, cloudinary_info, cloudinary_get_url, cloudinary_upload.
Does it support both Cloudinary folder modes?
Yes, it automatically detects whether your account uses dynamic folders or the older fixed folders.
Is there a safety mechanism for destructive actions?
Yes, destructive tools like delete or move return a preview with a confirmation token that must be used to finalize the change.
Can I use it for automated scripts?
Yes, it offers a scriptable CLI with JSON output and meaningful exit codes for CI/CD integration.