rclone-mcp-server
Provides tools for managing cloud storage remotes and performing file operations (copy, sync, list, mkdir, delete, etc.) via the Rclone RC API.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@rclone-mcp-serverWhat files are in my backup remote?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
rclone-mcp-server
MCP (Model Context Protocol) server for the Rclone RC API. Gives AI assistants the ability to manage cloud storage remotes, copy/sync files, list directories, and more — all through natural language.
Tools are auto-generated from the rclone-openapi spec using the rclone-sdk client. 98 endpoints, organized into selectable toolsets.
Prerequisites
A running rclone remote control daemon:
rclone rcd --rc-no-auth
# or with auth:
rclone rcd --rc-user=admin --rc-pass=secretRelated MCP server: rclone-mcp
Installation
All configurations below assume a running rclone daemon (see Prerequisites). Pick one transport — it decides the shape of your client config:
stdio — the MCP client spawns a local server process itself (
npx/node, or wrapped in a Docker container) and talks to it over the process's stdin/stdout. This is what Cursor, Claude Desktop, and opencode use locally.Streamable HTTP — a standalone server runs somewhere and clients connect over HTTP to its
/mcpendpoint. No client-side process is spawned.
Stdio transport (local processes)
Directly with npx (requires Node.js)
Cursor / Claude Desktop (.cursor/mcp.json or claude_desktop_config.json):
{
"mcpServers": {
"rclone-mcp-server": {
"command": "npx",
"args": ["-y", "rclone-mcp-server"],
"env": {
"RCLONE_URL": "http://localhost:5572"
}
}
}
}opencode (~/.config/opencode/opencode.json, or a project-level opencode.json):
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"rclone-mcp-server": {
"type": "local",
"command": ["npx", "-y", "rclone-mcp-server"],
"environment": {
"RCLONE_URL": "https://rclone.example.com"
}
}
}
}opencode's MCP config differs from other clients — do not copy the mcpServers
block above verbatim:
type: "local"is required; opencode uses it to decide how to spawn the server.commandmust be an array of strings; there is no separateargskey.Environment variables go under
environment, notenv.Use
"enabled": falseto disable a server.Restart opencode after editing — config is only loaded at startup. Its tools then appear under the
mcp__rclone-mcp-server__prefix.
With HTTP Basic Auth (rcd behind a reverse proxy). The same shape works for any
stdio launch — put the auth variables under env for Cursor/Claude Desktop as shown,
under environment for opencode, or as -e flags for Docker:
{
"mcpServers": {
"rclone-mcp-server": {
"command": "npx",
"args": ["-y", "rclone-mcp-server"],
"env": {
"RCLONE_URL": "https://rclone.example.com",
"RCLONE_USER": "your_user",
"RCLONE_PASS": "your_password"
}
}
}
}Via Docker (container over stdio)
Still stdio — the container just runs the same server
(ENTRYPOINT node dist/index.js, default stdio command) and the client starts it
with docker run. Use this when the client machine has Docker but you don't want
Node.js/npm installed there.
Step 1 — build the image (once, on the machine that runs Docker):
cd rclone-mcp-server
docker build -t rclone-mcp-server .Step 2 — configure the client to spawn docker run -i --rm ... rclone-mcp-server
instead of npx.
opencode:
{
"mcp": {
"rclone-mcp-server": {
"type": "local",
"command": [
"docker", "run", "-i", "--rm",
"-e", "RCLONE_URL=http://host.docker.internal:5572",
"-e", "RCLONE_USER=your_user",
"-e", "RCLONE_PASS=your_password",
"rclone-mcp-server"
]
}
}
}Cursor / Claude Desktop:
{
"mcpServers": {
"rclone-mcp-server": {
"command": "docker",
"args": [
"run", "-i", "--rm",
"-e", "RCLONE_URL=http://host.docker.internal:5572",
"-e", "RCLONE_USER=your_user",
"-e", "RCLONE_PASS=your_password",
"rclone-mcp-server"
]
}
}
}Notes:
-iis mandatory: it pipes the client's stdin/stdout into the container, which is how stdio MCP works.host.docker.internalresolves to the machine running Docker. Use it whenrcdruns on that same machine; otherwise setRCLONE_URLto the real daemon address. Note: Docker Desktop (Windows/macOS) adds this host automatically, but plain Docker Engine on Linux does not — there you must also pass--add-host=host.docker.internal:host-gateway. On Linux you can instead just setRCLONE_URLto the host's LAN IP and skip the flag.Environment variables must be passed with
-e(or--env-file) — a client's ownenvironment/envblock does not reach inside the container.Tune toolsets/read-only with e.g.
-e RCLONE_TOOLSETS=default,jobsor-e RCLONE_READ_ONLY=1(see Environment Variables).
Streamable HTTP transport (standalone server)
The server runs on its own and clients connect over HTTP — no client-side spawn. Use for remote hosting or web-based MCP clients that can't spawn local processes, or when you want one server shared by many clients.
Step 1 — start the server on a machine that can reach the rcd daemon. The rclone connection settings come from that machine's environment:
RCLONE_URL=https://rclone.example.com \
RCLONE_USER=your_user \
RCLONE_PASS=your_password \
npx rclone-mcp-server http --port 3000(KEY=value prefixes are bash/POSIX syntax — on Windows cmd use set "RCLONE_URL=...",
on PowerShell $env:RCLONE_URL="...".)
It listens on http://0.0.0.0:3000/mcp (the /mcp path only).
Step 2 — point clients at the endpoint instead of a local command.
opencode (~/.config/opencode/opencode.json):
{
"mcp": {
"rclone-mcp-server": {
"type": "remote",
"url": "http://localhost:3000/mcp"
}
}
}Cursor (.cursor/mcp.json, UI: Type = streamableHttp):
{
"mcpServers": {
"rclone-mcp-server": {
"url": "http://localhost:3000/mcp"
}
}
}Claude Desktop (claude_desktop_config.json) does not accept url entries (they are
silently dropped), so bridge over stdio with mcp-remote:
{
"mcpServers": {
"rclone-mcp-server": {
"command": "npx",
"args": ["-y", "mcp-remote", "http://localhost:3000/mcp"]
}
}
}Note: replace localhost with the server's hostname/IP when clients run on other
machines. The server binds all interfaces by default — put it behind a reverse proxy
with authentication (HTTPS) before exposing it publicly, and point clients at
https://your-host.example.com/mcp.
Also note that RCLONE_USER/RCLONE_PASS only protect the link between this server
and rcd — the MCP /mcp endpoint itself has no authentication, and the default
toolset includes write operations (copy/mkdir/delete). Do not expose it to untrusted
networks without the reverse-proxy auth above.
Configuration
These settings configure the server process itself. They work identically no matter which of the two ways the process gets started:
spawned by your MCP client — the client launches the server for you using one of the configs from Installation (
npx/docker, stdio mode). You don't run anything by hand; the settings live in the client's ownenv/environmentblock (or in-eflags for Docker).started manually in a terminal — required for the standalone
httpmode and for direct testing. You type the command yourself and pass the settings as shell environment variables or CLI flags.
Both channels feed settings into the same process. The two sections below document those inputs (environment variables and CLI flags) once.
Environment Variables
Variable | Description | Default |
| rclone RC daemon URL |
|
| HTTP Basic Auth username | — |
| HTTP Basic Auth password | — |
| Comma-separated toolset list |
|
| Set to | — |
| HTTP listen port for the |
|
CLI Arguments
The server's own command-line interface, used when you start it manually:
stdiois the default command — it is exactly the process a client spawns for you in stdio mode (Stdio transport), so you rarely type it yourself.httpis the standalone mode that must be started manually — theStep 1of Streamable HTTP transport.
rclone-mcp-server [command]
Commands:
rclone-mcp-server stdio Run with stdio transport (default)
rclone-mcp-server http Run with Streamable HTTP transport
Options:
--toolsets Comma-separated list of toolsets
--read-only Only expose read-only tools
--port HTTP port (http command only, default: 3000)Toolsets
Tools are grouped by API path prefix. Enable only what you need to keep the tool list focused.
Toolset | Paths | Default |
|
| Yes |
|
| Yes |
|
| Yes |
|
| No |
|
| No |
|
| No |
|
| No |
|
| No |
|
| No |
|
| No |
|
| No |
|
| No |
|
| No |
|
| No |
|
| No |
|
| No |
|
| No |
|
| No |
Special values:
default— the three default toolsets (core, config_read, operations)all— every toolset
Examples
Manual terminal invocations of the flags above (the manual channel — run the server
directly, e.g. for testing, instead of having a client spawn it). The same settings,
when the server is spawned by a client, go under RCLONE_TOOLSETS / RCLONE_READ_ONLY
in the client's env/environment block (or as -e flags for Docker):
# Default toolsets (12 tools)
npx rclone-mcp-server
# Everything (98 tools)
RCLONE_TOOLSETS=all npx rclone-mcp-server
# Just file operations and config
npx rclone-mcp-server --toolsets operations,config_read
# Default + mount
npx rclone-mcp-server --toolsets default,mount
# Read-only mode (no copy, delete, sync, etc.)
npx rclone-mcp-server --read-onlyMCP Resources
Beyond the call-based tools, the server exposes read-only resources so clients can pull context like a file URL. They are always registered (regardless of toolset) and are safe to enable in read-only mode.
Resource URI | Description |
| List of configured remote names |
| Full config dump with secrets redacted (pass, token, key, auth, …) |
| Running Rclone engine / Go version |
| Real-time transfer & job stats |
| Active background jobs |
| Directory listing or file content for a remote path (stat → list / cat; binary or >1MB files return metadata only) |
Examples:
rclone://remotesrclone://core/versionrclone://my-remote/— list the root of remotemy-remoterclone://my-remote/Books/guide.pdf— try to read a file
Read-Only Mode
When --read-only or RCLONE_READ_ONLY=1 is set, only non-mutating tools are registered. This excludes operations like file copy/move/delete, sync, directory creation, mount/unmount, etc. Useful for giving AI assistants safe, read-only access.
Usage Scenarios
The scenarios below build on the base installs from Installation:
use the stdio command/args block you picked there and only swap the env
fields. For the stdio examples below we use the published package via npx;
if you built locally, replace "command": "npx", "args": ["-y", "rclone-mcp-server"]
with "command": "node", "args": ["/path/to/rclone-mcp-server/dist/index.js"].
The examples use the Cursor / Claude Desktop mcpServers form. For opencode,
keep the server entry as an opencode local config and move every env key under
environment — see the opencode variant of Scenario 1 below.
Scenario 1 — Standard developer config (default tools + job monitoring)
{
"mcpServers": {
"rclone-mcp-server": {
"command": "npx",
"args": ["-y", "rclone-mcp-server"],
"env": {
"RCLONE_URL": "http://localhost:5572",
"RCLONE_TOOLSETS": "default,jobs"
}
}
}
}For a remote daemon behind a reverse proxy with HTTP Basic Auth:
{
"mcpServers": {
"rclone-mcp-server": {
"command": "npx",
"args": ["-y", "rclone-mcp-server"],
"env": {
"RCLONE_URL": "https://rclone.example.com",
"RCLONE_USER": "your_user",
"RCLONE_PASS": "your_password",
"RCLONE_TOOLSETS": "default,jobs"
}
}
}
}Same scenario in opencode form (~/.config/opencode/opencode.json):
{
"mcp": {
"rclone-mcp-server": {
"type": "local",
"command": ["npx", "-y", "rclone-mcp-server"],
"environment": {
"RCLONE_URL": "https://rclone.example.com",
"RCLONE_USER": "your_user",
"RCLONE_PASS": "your_password",
"RCLONE_TOOLSETS": "default,jobs"
}
}
}
}2 — Read-only (viewing only)
{
"mcpServers": {
"rclone-mcp-server": {
"command": "npx",
"args": ["-y", "rclone-mcp-server"],
"env": {
"RCLONE_URL": "http://localhost:5572",
"RCLONE_TOOLSETS": "default",
"RCLONE_READ_ONLY": "1"
}
}
}
}3 — Enable bulk sync + share links
{
"mcpServers": {
"rclone-mcp-server": {
"command": "npx",
"args": ["-y", "rclone-mcp-server"],
"env": {
"RCLONE_URL": "http://localhost:5572",
"RCLONE_TOOLSETS": "default,jobs,sync,sharing"
}
}
}
}4 — Everything (98 tools)
{
"mcpServers": {
"rclone-mcp-server": {
"command": "npx",
"args": ["-y", "rclone-mcp-server"],
"env": {
"RCLONE_TOOLSETS": "all"
}
}
}
}Common Tool Examples
The registered tool names are snake_case versions of each RC endpoint, e.g. /operations/copyfile → operations_copyfile. Since this is often an MCP bridge
to a remote rcd daemon, remote names must carry a trailing colon and there is
no local: remote on the daemon — use a real configured remote (e.g. my-remote:).
Discover (always list first, never guess a path)
{ "fs": "my-remote:", "remote": "" } // `rclone_lsjson` — root
{ "fs": "my-remote:", "remote": "Books", "recurse": true } // recursive
{ "fs": "my-remote:", "remote": "Books/x.pdf" } // `operations_stat` — one file
{ "fs": "my-remote:" } // `operations_size` — totals
{ "fs": "my-remote:" } // `core_about` — quota/capacitySingle-file operations
# copy one file
{ "srcFs": "my-remote:", "srcRemote": "backup/a.txt",
"dstFs": "other-remote:", "dstRemote": "incoming/a.txt" } // operations_copyfile
# move one file (source removed on success)
{ "srcFs": "my-remote:", "srcRemote": "a.txt",
"dstFs": "my-remote:", "dstRemote": "archive/a.txt" } // operations_movefile
# create / delete a directory or file
{ "fs": "my-remote:", "remote": "new-folder" } // operations_mkdir
{ "fs": "my-remote:", "remote": "tmp/a.txt" } // operations_deletefileDeleting a directory needs the
operations_advancedtoolset. The default tools only includeoperations_deletefile, which removes a single file — it cannot delete directories. Removing empty directories / a directory tree is done byoperations_rmdir/operations_rmdirs/operations_purge, which are grouped underoperations_advanced(default: false). To enable directory deletion, setRCLONE_TOOLSETS=default,jobs,operations_advanced. (mkdiris in the defaultoperationstoolset, so creating directories works out of the box; only deleting them requires the extra toolset.)
Bulk tree operations — always use _async: true
Whole-tree copies and syncs can exceed the request timeout. Set _async: true
so the daemon returns a jobid immediately, then poll job_status.
# copy tree src -> dst (additive, no deletes)
{ "srcFs": "source-remote:", "dstFs": "my-remote:", "_async": true } // sync_copy
# mirror dst to src (DESTRUCTIVE — deletes extra dst files)
{ "srcFs": "my-remote:", "dstFs": "source-remote:", "_async": true } // sync_sync
# poll the job
{ "jobid": 17 } // job_status -> until `finished` is trueLong-running tools that accept _async: sync_copy, sync_move, sync_sync,
operations_copyfile, operations_movefile, operations_size, operations_purge,
operations_delete, operations_copyurl, operations_check.
Share links (sharing toolset only)
{ "fs": "my-remote:", "remote": "share/folder" } // operations_publiclink
{ "fs": "my-remote:", "remote": "share/folder", "unlink": true } // removeLicense
MIT
This project is a secondary refactor based on rclone-ui/rclone-mcp.
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- Alicense-qualityAmaintenanceEnterprise-grade MCP server integrating Claude AI capabilities, enabling AI-powered conversations, file operations, and system tools via the Model Context Protocol.1Apache 2.0
- AlicenseAqualityFmaintenanceMCP server for the Rclone RC API. Gives AI assistants the ability to manage cloud storage remotes, copy/sync files, list directories, and more — all through natural language.564511MIT
- Alicense-qualityDmaintenanceAn MCP server with real AI capabilities (OpenAI/Anthropic) for natural language understanding, multi-step planning, and autonomous task execution, enabling intelligent file analysis, weather-based planning, and more.270ISC

@krovacloud/mcpofficial
Alicense-qualityAmaintenanceMCP server for interacting with the Krova Cloud API, enabling AI assistants like Claude to manage cloud resources.MIT
Related MCP Connectors
MCP server for AI dialogue using various LLM models via AceDataCloud
Cloud-hosted MCP server for durable AI memory
Hosted MCP server connecting claude.ai, ChatGPT and other AI apps to your own computer
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/angenge/rclone-mcp-server'
If you have feedback or need assistance with the MCP directory API, please join our Discord server